AI's energy appetite has Taiwan reconsidering the nuclear option The global surge in AI is placing unprecedented pressure on energy resources, with chipmakers such as TSMC consuming vast amounts of electricity to meet growing demand for advanced silicon. In response, Taiwan's government is signaling a potential shift in its longstanding opposition to nuclear energy to address its mounting …
@AC
"Small island with not much spare land appropriate to use for wind and solar"
Thats ok, there is little land and lots of water so off shore wind!
"Nuclear is the compact option."
Considerably. Much more energy generation for so much less space and its also reliable.
*Not the downvote
If the whole Asia would install insulated glass windows (also called double-pane) with proper gaskets, stop using resistive heaters when there are A/Cs (that can double as heat pumps) pretty much in every room and, dog forbid, insulate the buildings (at least the new ones FFS!) the current energy production would be enough for AI, bitcoin mining and a bunch of other stuff too
Similar comments are made frequently here in the northeastern US. We have a lot of old, drafty housing, built back when oil was cheap, and we have higher energy costs than most of the rest of the country. It's common to find that a few bucks spent on better insulation, windows, etc. is a (much) better investment than adding new supply. Even though energy production is well-subsidized in various ways, it's still often cheaper to avoid consuming a kilowatt-hour than it would be to generate said kilowatt-hour.
For other parts of the US with more modern housing and/or lower energy prices, the cost/benefit analysis can go the other way. Dunno about Asia in general, but can easily believe you've seen specific parts thereof where energy conservation would be the cheap way to go.
Um. I think you'd have to take it on a case-by-case basis.
My mother lives in a house built circa 1860, in an era that didn't insulate much. When her furnace died, the folks installing the new furnace evaluated the existing leakage with a thermal camera and a blower door. That showed that adding attic insulation and sealing around a bulkhead and a few other improvements would result in needing a (much) smaller furnace with lower fuel consumption.
That was a case where there was plenty of low-hanging fruit. If they did that in my house (built in 2006 and fairly tight to the weather), they wouldn't find many conservation-oriented improvements. Here in Maine, my mother's situation is probably more common than mine. I've no idea what the balance would be in northern China or Japan or Korea or... but would assume they have at least some old houses, too.
The right way to go here?
Are we at the beginning of a global revolution in technology with AI?
Is it so good, that we just say okay to submitting to a never ending power demand, by this tech, for this tech?
Anyone remember the Bitcoin mining issue as an example?
Rather than just keep throwing money and energy at AI, shouldn't we look at the best way to reduce power demand, or provide renewable supplies?
I thought about this. Power is a national resource for countries and yes they create their own but do we really want that? My understanding is that building the model costs more energy than running the model. e.g. ChatGPT-4 was 100days at 50Gwh to build though I don't know the cost to run.
What if governments passed regulations so that the building of these models is done on a shared basis? A non-profit entity that they all contribute to solely for the purpose of building the models. Sure there are going to be costs to run it once built and they will need their own setups for that but if done right there wouldn't be a cost excess involved for any of the companies and we get to be partially eco friendly in the process. You could take it one step further and get cross country cooperation and stick it somewhere cold with it's own dedicated undersea cables and build a couple of nuclear power plants.
A further thought is that if Taiwan think nuclear power's needed to cope with AI, they may have left it a bit late. Given that this is a politician's intention and there's no local nuclear power history, it'll likely be at least six years before construction can start, with commercial power export ten years after that even if they involve somebody who does have the expertise like Kepco. Whether AI is still a thing by 2033 who knows - and I'd imagine by then that nobody will accept the ludicrous energy demands the technology seems to get away with now.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
