home tags events about login
one honk maybe more

benjojo replied 28 Aug 2026 10:25 +0000
in reply to: https://social.treehouse.systems/users/dee/statuses/117172637692719907

@dee The water argument drive me nuts as well, like yes you might not want to build such a cooling system in Arizona, but when it's a race between (in a lot of situations) "consume a decent multiplier more power" or "add some water to the water cycle"... bleh. people.

I suppose a lot of this is just the increase in general skepticism in society on everything (certainly in the US), people are generally not cartoon villains and do things because they kind of make wider sense

benjojo replied 28 Aug 2026 10:22 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/8p6fB3lH1QF3bN23x6

Maybe it always ends up being this way when you understand the subject matter in a better detail, but the data center "debate" has for me obliterated a huge amount of credibility in quite a few people for making crappy and/or mostly uninformed arguments.

Certainly least in the UK it seems to have imported quite a lot of the American backlash (which reading into it seems to have some genuine merits, the developers of those projects are actually seemingly being bastards), but the UK isn't really the one ending up with those ""gigawatt"" data center projects.

It's my understanding that if the data centers were better located geographically it would probably be quite a nice idea to do a lot of AI training in the UK, we end up curtailing quite a lot of wind power anyway due to transmission capacity shortcomings so there's a lot of energy on the table that just literally cannot be used, it's not like the training isn't going to happen at this point instead it just happens in US/Malaysia where they burn a lot of coal and gas to make it happen instead

benjojo replied 28 Aug 2026 10:32 +0000
in reply to: https://mamot.fr/users/nemobis/statuses/117172654928802695

@nemobis I think the argument the you're going for here is probably separate to the one that I am making.

I am running on the assumption that the demand (for capacity) is there (at least the AI people are building it) and will probably happen wherever it is can be done.

(I do believe that a large amount of the AI build out is silly)

But the hardware/buildings are going to exist regardless, my argument is about where these things a built and run, I would rather have these sites exist in the UK where the power is comparatively clean than Malaysia or many US states.

It's also important to understand the quite a lot of these DC sites have nothing to do with AI, they are just normal carrier/colo sites that either replacing buildings that have reached the end of their life, or yeah, the demand for compute is generally increasing even outside of AI

benjojo replied 28 Aug 2026 10:47 +0000
in reply to: https://mastodon.social/users/kbm0/statuses/117172696802259942

@kbm0 @nemobis I see your point, but I also strongly believe that the hardware will exist at this point regardless (assuming everyone doesn't go bankrupt in-between the build out and the first day of real operation), as everything is very much ordered out already, hell, this is the core reason why SSDs and RAM are so expensive.

A lot of the AI/LLM workload (especially training) is basically bottlenecked on "Can I get a lot of power as soon as possible and as a bonus reasonably cheaply?", and less so about where it physically is (nothing in these workloads are latency sensitive), Running the LLMs is only lightly location dependent because of contractual data processing agreements.

I don't know the carbon economics of concrete and steel, but I assume UK concrete and steel is about as bad a Malaysian concrete and steel

kbm0@mastodon.social replied 28 Aug 2026 10:29 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/t59lmrNl4d8Vz24z9X

@benjojo The problem here is the commercial aims of the AI "industry" and their marketing, which has forced a diversity of things to be lumped into one basket. We could probably keep and develop a lot of the useful models we see being used in specific scientific and medical applications, and these might be viable with fairly modest computing facilities such as you find at a large university. But we are being invited to swallow the entire industry whole.

tmcfarlane@toot.comm.. replied 28 Aug 2026 10:51 +0000
in reply to: https://mastodon.social/users/kbm0/statuses/117172665776455004

@kbm0 @benjojo most of those scientific and medical models have nothing to do with the current AI hype cycle either.
You /can/ train your cancer cell recognition models in these facilities, but people were already doing that elsewhere. That's not what the build out is really for.
What does need the scale is ridiculous image and video generation. And hooking every single language based thing in to it for the sake of it.
These DCs aren't getting us new medicine.

tmcfarlane@toot.comm.. replied 28 Aug 2026 10:43 +0000
in reply to: https://benjojo.co.uk/u/benjojo/h/cG23wh87z222DfBMR3

@benjojo right. I've worked in DCs all over the country and they were almost entirely on industrial estates. The only exception really is Canary Wharf. When those building were put up they couldn't give the space away for free. No one wanted to live there, and no one wanted an office there. Putting DCs in them made an awful lot of sense. It's always wild being in those incredibly plush buildings with a those fancy foyers!

benjojo replied 28 Aug 2026 10:34 +0000
in reply to: https://toot.community/users/tmcfarlane/statuses/117172672157419230

@tmcfarlane The water talking point is basically moot, I've had some absolutely insane conversations where people are running under the assumption that the data centers are somehow "destroying" the water permanently somehow.

People are generally also not aware of the absolute comparatively insane water consumption rate that most thermal electricity generation plant has

benjojo replied 28 Aug 2026 10:36 +0000
in reply to: https://chaos.social/users/russss/statuses/117172682849398083

@russss sure, but even so I guess the UK power grid is far less awful on a kwh/co2 basis than most of the places these sites are being built instead.

Obviously the downside to doing it in the UK is that when the wind is good, power is super clean and cheap, when it's not I assume the viability gets questionable fast