Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- Rhetoric meets reality.
Texas rolled out the red carpet to corporate interests --- and then discovered that their electrical grid operation is totally inadequate to deal with the impact.
This is just one isolated example of a much larger issue related to the "AI race".
AI is power intensive and the "winner" of the "AI race" is likely be whoever can produce lower cost power and lots of it.
by jqpabc123 - This entire essay is so obviously Claudeby newsy888
- I don’t want to be hyperbolic, but to me this technological revolution more than any other in history feels like Silicon Valley and its VC class going to war with rest of human society.
> AI is needed to reindustrialize America.
It seems like they have no idea or are willfully deceiving why Americans want to reindustrialize USA. It’s not because they feel bad that things aren’t made in USA anymore, it’s because they want those manufacturing jobs back. If you replace humans with robots, it’s almost like a monkey paw’s fulfillment of a wish. They want to build 2 Trillion of data centers to get rid of your white collar jobs and your blue collar job, and you’ll be happy for it cause GDP go up. SMH
by sashank_1509 - the TLDR is that the current historic peak supply of the texas grid is around 100gw. The queue of datacenter demand is 460gw.
If the new capacity uses gas to deliver the power it will consume more gas than the USA exported last year.
So basically it will drive up costs _globally_ to a comic scale, for limited return (its already expensive to train new models, this will make it ruinously expensive to both train _and_ run)
by KaiserPro - I think data centers and power are the dark-fiber of this current "bubble".
Inference is 2/3rds of the power used in AI, and I expect the majority of inference to go local in the next 2-3 years, about the time these data centers will come online.
The other 3rd is training, and the major labs are training 6-7 times a year right now. But I suspect we will see a few things. 1) training iterations will slow as the improvements between training runs gets lower and lower. Is it worth spending an extra $500M(ish) for the small improvement in quality? Who is paying for that?
2) Training improvements and improvements in GPUs will require less power. Along with fewer training runs, as we add more and more tokens, the amount of power needed for a training run will likely asymptote at some point.
by pedalpete - FYI the Texas grid produces the most power of any state, twice as much in fact as either California or Florida.by seereadhack
- I don't think the model of highly concentrated data centers makes a lot of sense. From a security standpoint, you can target them easily. It's better for everyone (except perhaps a handful of elite power-mongers) if the hardware is distributed among the populace. I've pushed for something like this before, where each home has a 128GB+ powerful computing rig they can use to run LLMs, generate images and whathaveyou.
I get why the people in power want it, you can concentrate dependence on your asset, do surveillance and industrial espionage. But it's so obviously better to have the power distributed out among the people that it's a no brainer for the average joe/jane.
No doubt, there is still a need for large datacenters to generate models, but we should be looking to a more sustainable/distributed approach when it comes to inference, or running the models. That way people can keep their data, not have their ideas ripped off, not be subject to the whims of whatever policy some technocrat dreams up and not be shit out of luck if someone strikes us-east1 or some such place.
by state_less - 474 GW!!! Good God that's at LEAST the 5x our peak winter demand for the entire state.
And they want to mostly power it with on site GAS?!
What is even happening.
by Rapzid