AI Will Suck Up 500% More Power in UK in 10 Years, Grid CEO Says (bloomberg.com) 50
Electricity demand from UK data centers will jump sixfold over the next 10 years as a boom in AI requires increased computing power, according to the head of National Grid. From a report: That will ramp up pressure on the country's electricity network, which must move vast quantities of renewable energy from as far away as Scottish wind farms to data centers around London. And it's a grid already under strain from the accelerating electrification of home heating, transportation and industries.
"Future growth in foundational technologies like artificial intelligence and quantum computing will mean larger-scale, energy-intensive computing infrastructure," National Grid Chief Executive Officer John Pettigrew said Tuesday at a conference in Oxford. It's an outlook replicated in many other countries, which are grappling with how to fund the massive spending required to expand capacity. Global electricity demand from data centers, AI and cryptocurrencies may more than double over the next three years, according to International Energy Agency forecasts.
"Future growth in foundational technologies like artificial intelligence and quantum computing will mean larger-scale, energy-intensive computing infrastructure," National Grid Chief Executive Officer John Pettigrew said Tuesday at a conference in Oxford. It's an outlook replicated in many other countries, which are grappling with how to fund the massive spending required to expand capacity. Global electricity demand from data centers, AI and cryptocurrencies may more than double over the next three years, according to International Energy Agency forecasts.
AI Will Suck (Score:2, Troll)
Re: (Score:3, Insightful)
AI will still be around? (Score:2)
Yup.
Sucks so much that in fact I'm wondering if CEO John Pettigrew isn't overly optimistic in expecting the current crop of "Just make the datacenter even larger and AGI will surely emerge out of the next 5.0 version of CludeGPT or BingBard or whatever..." money burning bullshit "start-ups" will still be around in 10 years.
On the other hand bullshit like Tesla's "Full Self Driving" has constantly been promised for "next year" for approximately the last decade, so it's not impossible that by 2034 the same
Re: (Score:2)
Re:Maybe use the bittorrent mining energy? (Score:4)
Bitcoin, not BitTorrent. BitTorrent is a peer-to-peer file-sharing protocol. Bitcoin is the cryptocurrency that uses obscene amounts of energy for nothing.
Re: (Score:1)
Deal with today's problems first... (Score:5, Insightful)
If they are worried about power consumption, they should be dealing with what is already certain: EVs will need more power than the UK can currently provide, plus building out their power lines, plus building out their charging stations.
Will data centers also need more power? Maybe. But they are a lesser problem, because they can be build where the infrastructure can support them. So two of the above three problems with EVs won't exist. Also, with custom-designed AI chips and ever better models, it is entirely possible that the power requirements will not be all that dramatic.
Re: (Score:1)
Re: (Score:1)
I don't see myself interested in getting any sort of EV for the next few decades....so, by staying ICE, I'll reduce the strain on the grid a bit longer till that infrastructure can all be sorted out...which is going to take quite awhile at the current rate of planning/creation/repair.
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
" EV drivers generally stop and charge when doing something else like lunch "
Awesome.
How does that work when EVERYONE has an EV, there aren't enough chargers to go around and the
charge times are long ?
Re: (Score:2)
If current trends persist (Score:2)
Re: (Score:2)
Well sadly the trend has been continuing for over a decade now, computers despite being a lot more powerful are also consuming far more power, especially GPUs - of course I'm talking about the top-end as supercomputers and large computing farms use the most advanced cpu's and especially gpu's.
Better than Bitcoin (Score:1)
Seriously, this is like complaining light bulbs will consume most of the electricity... Yeah AI uses electricity. So what? It's getting more and more efficient every day. Pretty soon you will have ChatGPT5 on your iPhone using as much power as the wifi.
Re: (Score:2)
It's the scale that matters. Ie, bitcoin doesn't use that much energy if it's just you. But set up giant coin mining sites then it's very big. Some sites have deliberately set up locations where energy is subsidized or cheap (raising prices for all users) because the cost of energy involved is immense.
Also they're doing calculations that aren't typical for data center computer; the data center is I/O heavy, whereas crypto is compute heavy. AI will be a mix of I/O (like a data center) but also heavy duty
Re: (Score:3)
Sadly not quite.
Although CPU's and GPU's are getting more efficient and crazy powerful, they are also coming up with new ways of increasing the amount of power usage for the top-end chips. Thus, Intel CPU's at the top-end have never been more energy inefficient. Same with GPU's, the top-end, utilised by bitcoin farms and AI, use enormous amounts of power and it's only increasing with every generation.
e.g. over a decade ago, the top-end Nvidia 8800 GTX had a TDP of 145 watts;
a decade later, the top-end GTX 1
Value for cost? (Score:5, Insightful)
I have yet to see ANY indication that the cost of AI in energy, in natural resources, and in the pollution which results from both of these, will be lower than the benefits provided by the technology. Furthermore, it seems that at least in the short term AI will have significant social costs - increased loss of privacy, major labour market disruptions, further increases of wealth concentration, etc.
While one might argue that AI could help us fight climate change, it seems like a lousy bet to be further increasing greenhouse emissions for with a sketchy justification that we may solve some of that problem as a byproduct of AI implementation.
Our house is on fire, and we keep coming up with neat, cool, gee-whiz tech which not only distracts us from fighting the fire, but also is making the fire burn hotter and faster.
Re: (Score:2)
I have yet to see ANY indication that the cost of AI in energy, in natural resources, and in the pollution which results from both of these, will be lower than the benefits provided by the technology.
The best measure of a "benefit" is how much people are willing to pay.
If people won't pay for AI, it won't grow and won't be a problem.
Or, perhaps you are saying you are smarter than "common people," and you get to decide how they spend their money.
Have moral compass, not dollar compass (Score:2)
Re: (Score:2)
Cocaine sounds like a hell of a benefit by your measures
If you want to spend your money on cocaine, that's none of my business, and my cocaine consumption is none of your business.
Re: (Score:2)
Re: (Score:2)
Cocaine has strong uses cases, it achieves the desired outcomes, and it has established and resilient production and distribution chains. If you're concerned about social wellness or societal and health impacts, you should probably worry less about cocaine and more about Hershey.
Re: (Score:1)
Yep, let's all as a society just move back to living in the caves.
This by nature enhances "work from home" as an option....so, that should help the transition away from all other modern conveniences.
Re: (Score:2)
AI helping fight climate change feels like a fantasy. I think some tech illiterate people are hoping that they can just build their own Oracle and then ask it hard questions and get good answers. Right now, AI is good at giving bad answers, AI is good at stitching together images and text and other things. It's not general intelligence, and if it were general intelligence it likely won't be as good as the average human for decades (or centuries), so why not just use an average human?
Yes, AI can do some t
If AI is so smart (Score:1)
Re: (Score:3)
Why canâ(TM)t it figure out how to reduce its power consumption
Machine learning is being used for new transistor designs and new PNR (Place and Route) algorithms for ASICs that will reduce the power consumption of future AI.
Machine learning is also being applied to deep learning hyperparameters and neural pruning that will reduce power consumption on existing hardware.
Pruning (artificial neural network) [wikipedia.org]
Hyperparameter (machine learning) [wikipedia.org]
or design new means to generate power?
Machine learning is being applied to semiconductor technologies used in PV solar and the design of turbine blades.
Re: If AI is so smart (Score:1)
Machine learning is old school stuff. We are primarily talking about LLM which has no true value except for the hype right now filling the pockets of Microsoft through its subsidiary OpenAI. We were doing auto-routing on PCB with âintelligentâ(TM) algorithms that had a function you could teach it specific routes to be preferred when I was in school, and that is a long time ago (20y+), back then it was super expensive, my EE school had 1 license on 1 486DX2 which was shared with the entire school a
Re: (Score:2)
Machine learning is old school stuff. We are primarily talking about LLM
LLMs are created with machine learning.
Re: (Score:1)
Sure, at the very core training happens through a form of ML but LLM are really large graph databases.
I know where the energy can be harvested (Score:2, Troll)
Stop the AI brainwashing and you'll save gigawatt-hours of websites not serving billions of pages talking about this stuff for nothing.
AI efficiencies? (Score:2)
Off-Cycle Increases - Be Careful (Score:5, Insightful)
Items like the electrification of home heating are something of a shell-game. Sure, they represent new drain on the electrical system, but they're a bit of a shell game on consumption. The usage was there - it's changing source. So consumption-wise it's flat-ish, with the "ish" part being about the relative efficiency of the act of heating the home. These things matter, and they're complicated... but they're reasonable.
AI consumption is brand new, unprecedented draw, without a historical parallel. So you'd better be damned sure that whatever you're doing with it s worth that cost. Running models improving medical treatments or, say, building better automated radiology assessments strike me as reasonable. Generating 300 candidate album covers with Midjourney is not.
This is a classic "just because you can..." moment. A lot of what we're doing with it is just plain stupid.
FFS, just ban it already. (Score:5, Interesting)
Also saves energy... (Score:2)
They can't measure how much energy is saved from having better information in many cases.
Mandatory Simpsons quote (Score:3)
"I predict that within 10 years, computers will be twice as powerful, ten thousand times larger, and so expensive that only the 5 richest kings of Europe will own them" [youtube.com]
- Dr Frink
If that's so... (Score:2)
...then they should be charged premium rates, and home accounts should be free.
Move the computing to the energy... (Score:5, Interesting)
So What? (Score:3)
Why is everything presented in panicked terms? Every goddamn sentence could be preceded with "oh dear God" comma.
It's because that's the only way to get anyone to click on it. It could all be a giant fragrant load of BULLshit, but that's not what's important.
The easiest way to test this is to add "please click" comma to the beginning of every breathless headline, if it fits, then its horseshit fictional clickbait. Thank you.
Possible solution (Score:2)
Since we know more power will be needed, perhaps we should start working on the problem now. Maybe use something like AI to work out possible avenues to approach the problem.
Oh wait.
Two things come to mind (Score:1)
1) Got nuclear?
2) Got coal?
Re: (Score:2)
AI Blockchain (Score:2)
AIs will store all their data in the blockchain. The awesome compute power of this combination will be used to find new ways to waste energy.