The word for countries that don't have lots of energy to use is "poor". That is the choice successive governments are making for us. It's not just AI, it's everything.
I think that's true to some extent, though of course we can sometimes try to do the same things more efficiently (use LED lightbulbs not fluorescents etc), so to me it's a question of what's by choice and what's being forced
The general pattern for rich countries who find a more efficient way of lighting their world is to have more lights.
Define "forced" I suppose, but I don't think many manufacturers have chosen to have electricity prices so high that they've had to shut down. I certainly don't remember a choice being given between a massively expensive, unreliable energy supply and continuing with any one of several alternatives.
Which reminds me of something I'd love to introduce: if you select a "100% renewable" energy provider then you only get energy that's generated from renewables. If its dark and calm, your power shuts down. It would, I think, rapidly clear up some differences between expressed and revealed preferences.
I agree up to a point, and certainly think that excessive prices in the UK are a problem. But there’s some interesting graphs on OWID for example - the US has had gently falling energy per person lately, and I don’t think all of that is down to outsourcing manufacturing https://ourworldindata.org/profile/energy/united-states
This is spot on, and compounded by the problem of grid connection in the first place. According to Tim Shipman’s recent piece in the Spectator, our hopes/claims to become a leader in AI are completely kiboshed by Net Zero and Miliband’s resistance to the data centres being built when the UK has limited grid capacity. According to Shipman:
“Miliband is the Mr Nyet of the AI Energy Council. ‘Every meeting is basically Ed saying: “It’s impossible. Can’t do it,”’ says a frustrated official. ‘He’s got his Green Energy 2030 target to hit and he won’t be pushed off it. He has a limited number of grid connections and doesn’t want to give priority access to a demand-side project like an AI data centre over a supply-side onshore wind farm.’
But I think it's a reasonable question to ask whether and why international companies will choose to host data centres in countries where they have to do this kind of demand management, as opposed to putting them somewhere like Spain or Morocco where solar plus batteries probably can deliver something close to 24/7 ..
Yup, definitely, it's an interesting question. Data connectivity, worker skills. Cheap reliable power in hot countries might be offset by higher AC bills and water consumption, I've got no idea. The 'just price carbon' response is quite attractive.
I've no evidence for this but mistaking the 5-yearly budget for a yearly budget in each of those five years sounds like the kind of mistake an AI tasked with producing the numbers for the report might make. That sounds more plausible than a human energy consultant making that mistake.
While we myopically build carbon budgets within our borders, data centres will simply be built in other jurisdictions. Global emissions will likely end up higher, but our politicians will proclaim our virtue and we will all end up poorer and more dependant on foreign regimes for our future economic development.
The calculation problem is why you should just price carbon per tonne at source and have done with it. All the time & effort put into making these report & administering the gate keeping around productive economic endeavour is functionally a dead loss. It should make you furious.
Some of the Planning Applications must be real, true. But it seems some of the "investments" are hype or, worse, some complicated tax dodge/subsidy claim.
The word for countries that don't have lots of energy to use is "poor". That is the choice successive governments are making for us. It's not just AI, it's everything.
I think that's true to some extent, though of course we can sometimes try to do the same things more efficiently (use LED lightbulbs not fluorescents etc), so to me it's a question of what's by choice and what's being forced
The general pattern for rich countries who find a more efficient way of lighting their world is to have more lights.
Define "forced" I suppose, but I don't think many manufacturers have chosen to have electricity prices so high that they've had to shut down. I certainly don't remember a choice being given between a massively expensive, unreliable energy supply and continuing with any one of several alternatives.
Which reminds me of something I'd love to introduce: if you select a "100% renewable" energy provider then you only get energy that's generated from renewables. If its dark and calm, your power shuts down. It would, I think, rapidly clear up some differences between expressed and revealed preferences.
I agree up to a point, and certainly think that excessive prices in the UK are a problem. But there’s some interesting graphs on OWID for example - the US has had gently falling energy per person lately, and I don’t think all of that is down to outsourcing manufacturing https://ourworldindata.org/profile/energy/united-states
This is spot on, and compounded by the problem of grid connection in the first place. According to Tim Shipman’s recent piece in the Spectator, our hopes/claims to become a leader in AI are completely kiboshed by Net Zero and Miliband’s resistance to the data centres being built when the UK has limited grid capacity. According to Shipman:
“Miliband is the Mr Nyet of the AI Energy Council. ‘Every meeting is basically Ed saying: “It’s impossible. Can’t do it,”’ says a frustrated official. ‘He’s got his Green Energy 2030 target to hit and he won’t be pushed off it. He has a limited number of grid connections and doesn’t want to give priority access to a demand-side project like an AI data centre over a supply-side onshore wind farm.’
Thanks, nice quote, I hadn't seen that!
It's not necessarily true that data centres are always 24/7 loads, the Economist cite Google and Microsoft both experimenting with load reduction/shifting in Indiana and Ireland https://www.economist.com/finance-and-economics/2026/03/05/americans-electricity-bills-are-up-dont-blame-ai
There was a National Grid trial on this too recently. https://www.nationalgrid.com/about-us/innovation/emerald-ai
Particularly for AI training workloads I'd assume it's reasonably trivial to pause these or scale them back to reduce power consumption on demand.
But I think it's a reasonable question to ask whether and why international companies will choose to host data centres in countries where they have to do this kind of demand management, as opposed to putting them somewhere like Spain or Morocco where solar plus batteries probably can deliver something close to 24/7 ..
Yup, definitely, it's an interesting question. Data connectivity, worker skills. Cheap reliable power in hot countries might be offset by higher AC bills and water consumption, I've got no idea. The 'just price carbon' response is quite attractive.
What stopping having these data centers in Iceland using geothermal power?
I'd say nothing at all, or similarly putting them in Spain or Morocco and using solar and batteries ..
I've no evidence for this but mistaking the 5-yearly budget for a yearly budget in each of those five years sounds like the kind of mistake an AI tasked with producing the numbers for the report might make. That sounds more plausible than a human energy consultant making that mistake.
I don't know for sure, but the report implies it was signed off by human beings, so you'd like to think they'd have spotted it!
The tragedy of the commons is a global one.
While we myopically build carbon budgets within our borders, data centres will simply be built in other jurisdictions. Global emissions will likely end up higher, but our politicians will proclaim our virtue and we will all end up poorer and more dependant on foreign regimes for our future economic development.
That's very true!
Oliver - I know you don’t like Bluesky, but a write-only account that posted links to your pieces would be useful.
Not sure that I see the point - if people on Bluesky want to see my stuff then they can just get it at source here
The calculation problem is why you should just price carbon per tonne at source and have done with it. All the time & effort put into making these report & administering the gate keeping around productive economic endeavour is functionally a dead loss. It should make you furious.
Maybe some of us like doing futile calculations!
NGH while the polluters have so much political influence. Tragedy of the commons as Oliver wrote.
Don't worry, there are no (new) datacentres bring built. It's all smoke and mirrors.
https://www.theguardian.com/technology/2026/mar/09/revealed-uks-multibillion-ai-drive-is-built-on-phantom-investments
OK, but I think that at least some of the data centres are real
Some of the Planning Applications must be real, true. But it seems some of the "investments" are hype or, worse, some complicated tax dodge/subsidy claim.