In our view, that’s a good thing. But as the anti-datacenter movement has grown, it’s come under fire from all sides, including from liberal critics who dismiss it as another privileged form of nimby (not in my backyard) politics with naive demands. A New York Times op-ed, for example, called the fight against datacenters a “myopic” “distraction” from the “real fight”. In truth, anti-datacenter organizing is the real fight, one centered on an industry choke point that people can reach out and touch. This brewing populist resistance isn’t just about limiting local development – it represents a critical new front in the fight against tech-enabled authoritarianism. Where else can people push back on job-eating algorithms, distorting deep fakes, and autonomous drone strikes?
We’ve found some trivial issue we can organise around and thereby haave comfy jobs sa anti-capitalists.
Well done, Lads
Why does the image of Wolfie Smith always burst into my mind when I read some shit from the Graun? Is it the childish, dopey. no nothing stench coming from each syllable?
Do you know, I think it might be…
critical new front in the fight against tech-enabled authoritarianism
As opposed to the analogue human authoritarianism to which these people subscribe.
“Where else can people push back on job-eating algorithms, distorting deep fakes, and autonomous drone strikes?”
Do people think stopping a datacentre from being built in Maine will stop that?
You can do the algos and deep fakes anywhere. Get a server from AWS, Azure, Alibaba anywhere in the world. And the autonomous drone strike stuff won’t be running on these. It’ll be on the department of war’s own servers in a basement of a government building.
Do people think stopping a datacentre from being built in Maine will stop that?
Yes, in the same way that idiotic middle-class wimmin and soyboys here think that destroying our economy with Net Zero will save the world.
These people really do think a hundred years before the times. Where there’s a demand someone, somewhere will want to supply. So if you’re opposed to local job eating algorithms you need to interdict the supply. Oppose 5g telcom, fiber optic cable & low cost bandwidth. Oh! Those are things they’re in favour of.
Everybody knows that 5G is used to control the nanobots in vaccines.
If “we” built a data centre that could use locally generated electricity and rely on plentiful supplies of fresh water the obvious spot would be somewhere on the fringes of the Highlands. Lots of offshore wind power thereabouts and lochs full of water. They could be sold the Cruachan pumped storage scheme to help them with still weather. And of course you could cover the buildings with solar panels.
How big a datacentre needs 440MW of electrical power?
impractically big….
Numbers differ, but the totality of official datacenters in Clogland get up to *about* half that in sheer power, for a consumption of roughly 5 GWh annually, about 5% of national electricity consumption.
You actualy *really* want one near any Nuke planned.. They make as good a stable base load as heavy industry, like….a steel or aluminium plant.. and, despite the moaning and screaming of the Greentards, are *far* less polluting.
Then again…GreenTards gotta ‘Tard…
There are basically 6 factors for data centres. Power, water, air temperature, distance, expertise and cost.
You need a steady power supply. They have backup but it doesn’t last that long.
Water. You need a bit. You don’t want one in really arid conditions but the talk about water is journalistic twattery.
Air temperature. Ideally, in a cool place, so the cooling of water and fans doesn’t work so hard.
Distance. Can’t be too far from users because of poor latency
Expertise. You need contractors who can come in and do installations and repairs.
Cost. Can you use cheap land? Can you get tax breaks.
Slough is quite a popular place. Near London, ticks for power, water, and there’s now lots of companies that can do the work and relatively cheap. Swindon has a few places. Amazon opened a centre up at Chiseldon which has the benefit of being at a high elevation, quite exposed so gets lots of free cooling. There’s some around Dublin, and they’re next to the Liffey, so the river helps to cool the building and tax breaks.
No-one is going to put them in the Highlands. Scores for being cold, land is cheap, but poor latency and where are you going to get people? And what happens if the Tartan Nazis get independence and decide it’s theirs now?
Do you really need good latency for AI data centres? I get it for anything that’s hosting gaming but a few extra milli seconds waiting for AI to respond is hardly going to break their business model?
Not for AI uses, not really. But I was being more general about data centre uses.
I wonder that the latency over a few hundred miles vs a few miles is very low, a few nanoseconds. And a few milliseconds is in fact a long time in computing.
You’d be able to “get people” who are part of the White Flight from Thirdworldistan.
This “water use” thing is bollox. We’ve been there a while ago. The figures were quoted, in litres, looked large because people don’t have any idea of scales. Annual usage of a data center as quoted, taken as a flow rate, were equivalent to leaving a single bathroom tap running.
And no one with any sense is going to be running tapwater through a cooling system. You’d calcite the fuck out of it. You use a treated volume of water which goes through a heat exchanger to re-cool it & recycle.
There are quite big differences depending on if the system is closed loop like you’re describing or open loop like evaporative cooling – uses much more water but lower on energy costs, which is why it’s been more popular than you might expect. Data centres in areas where water supply is expensive or unreliable can stick with air cooling but that increases electricity usage and struggles with the higher power of workflows like AI. Another zero-water solution is immersion cooling with dielectric fluid, much more efficient but it’s costly. Sort of game that Castrol are getting into for obvious reasons.
From a survey of data centres in England for techUK – not sure how representative it is – 51% use waterless cooling (I’m guessing not the ones doing high power work like AI), 64% use less than 10k cubic metres of water per year (the quoted comparison is “less than a typical leisure centre and similar to a Premier League Football Club”) and 4% use over 100k cubic metres per year (which is more industrial in scale). https://www.watermagazine.co.uk/2025/08/22/new-report-sheds-light-on-commercial-data-centres-water-usage-in-england/
In comparison average household water use in England is 136.5 litres per person per day according to https://www.gov.uk/government/publications/water-resources-2024-2025-analysis-of-the-water-industrys-annual-water-resources-performance/water-resources-2024-to-2025-analysis-of-the-water-industrys-annual-water-resources-performance which is 50 cubic metres per year. So a data centre using 10k cubic metres per year is the same as 200 people use domestically, and a large one using 100k is the same as 2000 people domestically. More than a tap left on but not absolutely enormous numbers.
As for a really big data centre: Project Marvel in Bessemer, Alabama was pitched as a $14 billion hyperscale data centre and the memo it sent the water company stated that it would require 2 million (US) gallons daily for cooling. If I did the conversion right that’s 2.76 million cubic metres per year, the same as tens of thousands of people. At that kind of scale you obviously do start to have an impact on the local community. https://alabamareflector.com/2025/11/19/bessemer-officials-approve-data-center-construction-despite-environmental-concerns/
Re the previous discussion about power, that one is slated for 1,200 megawatts of electricity consumption. For orders of magnitude that’s like MILLIONS of people. Something like 3% of UK electricity demand. So at least for this example it’s clear the energy issue is far more serious than the water. https://birminghamfreepress.com/2025/06/21/bessemer-officials-push-zoning-changes-for-massive-data-center-despite-widespread-opposition/
Annony. It’s about water use. If you want to do a lot of open cycle cooling, you choose somewhere you can abstract & after it does cooling return it to where you abstracted it. So there’s not any water actually being used. Given the cost of water, it’s difficult to believe anyone’s using great quantities of it & just pouring it down the drain. Even if they are, it’s not actually lost. It’ll end up as the flow in some river. It will contribute to the essentially required flow enabling someone else to abstract. But there’s so many easy & cheap ways of recycling.
As for the amounts of water “used”, the number I calculated from was quoted by one of these data center objectors. I used that to calculate the flow rate. To give you some idea of what that means, I used a similar flow to fill our swimming pool which held about 200m³. Took about 5 days non stop. Having a fair idea water inputs & outputs I’d guess you’re also probably talking about the usage of a medium sized office block. So one might comment that if data centers are going to replace jobs they’re going to be using a lot less water.
An open loop system with a cooling tower does consume a lot of water because the point is to evaporate it. You’re right there’s also (warm!) liquid waste water that can be returned to the system for treatment, so there’s a difference between water “withdrawn” and “consumed”. I’ve seen various figures quoted for data centres with cooling towers, ranging from 30%-80% of withdrawn water being evaporated away. Even at the lower end that’s a lot of water use. A lot of players in the industry are moving away from that model though.
Funnily enough, manufacturers of cooling tower solutions for data centres still claim it saves on water because of the greater energy efficiency – their argument is that if a less efficient method was used at the data centre, then the utility company would need to generate more electricity, and power stations also guzzle water.
You’re talking about this as if it’s some sort of unique problem. It’s exactly the same as if you build a power station. The process generates heat so you need to cool it. The simple answer is don’t put you’re data center somewhere where you have a limited supply of water. Put it by a river or a coast. Since power stations are around 50% efficient, 440kW would require the cooling of a medium sized 1MW version.
Of course heat in itself is a valuable commodity. We spend enough producing it. Why not use it?
What I do find amusing is the people who most bitch about this supposed problem are the perpetually online. On X & Farcebook & whatever. All run on datacenters
You’ve got a lot more choices for cooling a data centre than you do for a power station, so you don’t even have to avoid places with limited water. Power is by far the bigger problem. Open loop water cooling is just a nice solution because of how cheap and efficient it is.
But if the location merits it you can stick to other methods that use less water. Just need to suck up the higher costs and energy requirements. Sometimes you’ve got no choice, like data centres are clearly needed in Singapore but water there is scarce and tightly regulated, so the country has become a test-bed for alternative cooling methods. They also introduced a “Tropical Data Centre Standard” where you run hardware a few degrees C above usual recommendations to save water and energy. Doesn’t necessarily harm the hardware or performance but margin for error goes down. https://www.businesstimes.com.sg/companies-markets/telcos-media-tech/efficiency-initiatives-raise-data-centre-temperatures-leave
Air cooling is the traditional zero-water option provided the compute requirements aren’t too high power (i.e. less useful for AI workloads), and many big players like Microsoft are getting more into closed loop water cooling or – if they’re feeling fancy – immersion cooling. https://datacentremagazine.com/news/how-are-companies-pioneering-data-centre-zero-water-cooling
It is possible to do something useful with the waste heat – it’s low grade heat but it still gives something for a heat pump to work with. The ideal situation is heating dense residential or industrial buildings nearby. Countries like Finland with district heating systems have a headstart, but those systems are rare in the UK. Some new London data centres are having heat pumps installed as part of new district heating networks. Obvs still a waste of heat in summer. https://datacentrereview.com/2026/01/data-centres-dont-have-to-waste-heat-they-could-heat-our-towns/
Unfortunately it’s common for water utilities to only supply potable water – a wasteful level of processing given that (with the right treatment) you can use greywater or river water in suitable chillers. So in places where a data centre can’t abstract its own water and the water utility doesn’t pipe non-potable water to industrial users, the data centre is in competition with people for drinking water. This is very common and I think is why the water issue gets so much negative attention. A related legacy infrastructure issue is that even in places where hydrologically there’s no need to worry about water stress, utilities can struggle to supply new data centres due to limited pumping capacity. But ideally you’d prefer data centres there rather than a region undergoing water depletion which happens to have spare pumping capacity.
Another example of how the Grauniad means something different when it says “democracy”.
To them it doesn’t mean “rule by the people” – it means “rule by the right people (the ones who write for the Grauniad)”
OMFG. How does someone this stupid get a job?
Double OMFG! It’s a team!
Astra Taylor and Saul Levin
I’m guessing Astra’s favourite drink is a Mokka.
Similarly, I listened to a gushing report on the World Service on how “journalists can securely deposit digital reports in case they get killed/captured/etc” Wow! They’ve discovered servers! Only 70 or so years late.
Rage against the machine.
The Guardian article is disjointed. They correctly point out that AI is fallible. Yesterday, I asked AI a question. It rephrased my question, then answered it. Not what I asked. Useless waste of time. This the Guardian’s crack team’s reason why AI must be challenged. And ANY BLOKE should be able to stop it! Good luck with that fight.
“All computer results must be analyzed for reasonableness.” — Gamecock’s father, 1955
As far as AI’s fallibility, that will evolve out. Companies will find out how far they can take their trust in it. This is literally what has happened since the beginnings of computerization 70 years ago. I read a story a few weeks ago where AI completely erased a company’s entire database. A catastrophic AI blunder.
Their notion that the AI can be stopped is delusional. Delusional, a condition of employment at Guardian.
I’ve had the same trouble with GoogleTranslate. I use it for writing in Spanish because my language skills aren’t up to it. But I do have enough to write my English in the same form as Spanish, so I should get a clean translation. That used to work fine.
Some of the results recently have been a disaster. The output’s been nothing like the input. Words are not even slightly connected to ones I’ve used. Either it has no relevance to what I’ve written or it’s distorted to be the opposite. I presume it’s just churning out what other people have said in the similar context.
The general problem with AI is that it’s somewhat good, somewhat reliable. It’s not professionally good, though.
Google Translate is what I used holidaying in Lake Garda. Apart from being able to do a big of Italian so knowing if it gives me something utterly wrong, being a bit wrong when asking for food doesn’t matter that much. There’s another human who will try and help me, and we’ll figure it out.
I wouldn’t use Google Translate for translating a website because that’s how you get the odd comically bad thing. Or, the wrong word in context. I’d pay for a human being.
Most people wetting themselves about AI don’t get this. If you’re an amateur it looks great. It’ll write you some code, and it might work. But it might not. It might blow up on you in a couple of weeks.
There are some tricks to getting reliable machine translations. Trying to use the sentence constructions of the intended language I’ve mentioned. Avoiding sentences with subsidiary clauses. Punctuate! but don’t use commas in lists. Don’t do something that’s common & habitual in English & use trailing adverbs. Always use the pronouns, even if you normally wouldn’t, because you’re going into a gendered language. As in : “She walked down the road and then she got on the bus”. Or could end up with he got on the bus or you got on the bus.
It’s often a giveaway that machine translation is operating when it produces phrases like: “The piano slid from the stage and then he fell into the stalls.” It’s obvious to a fluent English speaker that ‘he’ should read ‘it’, but a LLM has no idea of context, it will just copy the most frequent translation it can find that matches the original text.
Try DeepL, its sound in German and from what I can gather French as well.
I second DeepL. Been impressed. It’s a tough job to get an idiomatically “natural” translation rather than an overly literal word-for-word one, but which also doesn’t skew your meaning or add its own inventions. DeepL seems to get the balance right.
Also improves your results if, when you want to translate your own words rather than someone else’s, you take care to write things without ambiguity. Beware of using words or phrases that might have multiple meanings and add a little clarification if necessary.
DeepL is excellent in Spanish too, I use it professionally- but only as an aid. Would never send a tranlation to a customer that I hadn’t checked thoroughly. Does save time though.
Thinking about what your wise father said (“All computer results must be analyzed for reasonableness” – yes) and also what Western Bloke said about it being good but not professionally good, what will AI actually be most useful for?
Using AI for something that you really don’t understand well enough to be able to check, especially when errors have high cost, is unwise. Don’t vibe code anything important. Don’t use it to translate business documents into a language you don’t speak. Don’t use it to perform calculations you don’t understand. Perhaps there are times when “it’s usually good enough” is fine for low stakes stuff (the example of translating a menu is a good one, summarising an unimportant document is another – rarely much point summarising an important document as there’s no time saving if you need to read thoroughly to check the summary anyway) but by definition there’s not much value there.
Some of the best added value is going to be areas where AI saves time doing something relatively menial but is straightforward, at least for a professional, to check the results. In principle that’s a productivity boost.
That might mean a developer writing some boilerplate code. For a doctor, a diagnostics “second opinion” from a medical scan. (If the model is good enough you might use 1 doctor + 1 AI when in the past you used 2 doctors.) A bilingual person using it to get a good first draft of a translation. Generating images as illustrations or design mockups. Rewriting text to change the style.
There’s clearly enough value there that it’s not going away. The potentially more transformative thing would be AI generating new knowledge not just remixing its training data. I know it’s touted for drug discovery, but then pretty much everything gets touted for drug discovery given all the dollars in it. AI-assisted chip design is another. Mathematicians have successfully chained AIs to proof assistants (which check the validity of a chain of reasoning so automates the “check for hallucinations” problem) and used it to find new proofs of conjectures. Security researchers have AI-written exploits for computer systems, these can be checked by running the exploit script and seeing if it successfully breaks in. But I don’t think there are many fields like this which lend themselves to automated verification of the AI output.
So absent that, and if you’re not going to have human verification either, then you really do need the fallibility to be evolved out before you can use it in a high-stakes area. Even for autonomous vehicles, a field that has been researched for decades and has had tens of billions of dollars thrown at it, that seems a tough nut. We clearly aren’t all about to become redundant to AI in the near future.
I don’t see autonomous vehicles as being a big problem, but not the way they’re doing it. They’d work with an interactive traffic management system when vehicles communicate with the system & each other. All entirely possible, albeit expensive to implement.
But also the metric shouldn’t be whether they’re perfect but whether they’re better than human drivers. And that isn’t asking a lot.
We”re having something relates to it here. Jan 1st we went over to putting V16 GPS enabled warning flasher on top of broken down cars instead of getting out & using a red triangle some distance up carriage way. They’ve had truck drivers not seeing the warning & wiping out stalled vehicles. First question, would they have seen the red triangle in time? Seconds, since the hazard can be localised within a couple of metres, why aren’t at least HGVs carrying a receiver would provide a warning? Bulk manufactured they shouldn’t cost much more than the 40€ V16s.
Yep, I’m surprised there’s not been more standardisation on communication between autonomous vehicles – be a big help if you could “see through” the vehicle in front because it shares its sensor input with you.
A long distance trucking route on an autonomous-only road would probably have been achievable years ago but the infrastructure isn’t worth building, especially if it’s only for a transitional period. As soon as you start adding pedestrians and human-controlled vehicles into the mix then it’s harder.
While I agree “as safe as human drivers” is the better and more achievable yardstick, there will undoubtedly be people (AI skeptic grifters and professional drivers/their union representatives) who push for stricter and more unrealistic criteria. Will be interesting to see who has the political power to win that battle – Big Tech has been spending a lot on lobbying and cultivating the fear of “losing the AI war to China” so they’re in with a decent chance.
No, Timmy, just no.
This isn’t about anti-capitalism, as you so quaintly put it, it’s about citizens having genuine concerns about the impact of such data centers will have on the quality of life in the areas where these data centers are to be built: Health and safety concerns, energy cost concerns, etc., etc., etc. And for what?
The reality of it is that AI is shit at the consumer level. It provides no material benefit the average citizen can actually discern. AI provides material benefits to government and large corporations. It is for the collection of large amounts of information for analysis and manipulation to benefit governmental and corporate actors. And when those actors are scumbags like Barack Obama, Chuck Schumer, Bill Gates and Mark Zuckerberg, scepticism, distrust and outright opposition constitute an understandable, logical and necessary set of responses.
Microsoft isn’t trying to shove Co-Pilot our throats because they think it will benefit the masses… If you look at the legal disclaimers Microsoft has for Co-Pilot, Microsoft states it is for “recreational use only”. Really? So why the push? Well, perhaps it is to cover for Microsoft’s attempt to build an AI regime for someone other than the average consumer. Perhaps it’s cover for Microsoft, Oracle and dozens of other companies and governments to quietly extend control over its citizens.
There are anti-capitalists involved in this, but they aren’t named Joe Jones or Bob Rogers, they are named Mark Zuckerberg, Larry Ellison, Bill Gates and Satya Nadella.
So fucking what? I don’t like the color of my neighbors car.
This ABSOLUTELY is anti-capitalism. Dennis the Missing, if that’s his real name, doesn’t want people to be FREE to build what they want. Cos reasons.
Is that Dennis the different name every time? Long time no see. And it makes no difference to Bill and Zuck if the data centre is in your backyard or in deepest Utah. It will still exist. In an anonymous building with no external indication of what it is. Like all the others you aren’t particularly aware of.
the anti-datacenter movement
What does it pay? I’ve got a couple of days a week I could protest, if the pay is good.