Thereβs an eye opening article in FT quoting Morgan Stanley about the size of the generative AI bubble developing.
https://www.ft.com/content/7052c560-4f31-4f45-bed0-cbc84453b3ce
βHyperscaler funding of $300bn to $400bn a year compares with annual capex last year for all S&P 500 companies of about $950bn.β
Paywallless: https://archive.ph/2025.07.30-135255/https://www.ft.com/content/7052c560-4f31-4f45-bed0-cbc84453b3ce
Tl;dr - if GenAI doesnβt pay off - as Morgan Stanley point out, it may not meet expectations - it will be 10 times higher than the previous bubble pop. π₯€
πΏ
@GossiTheDog@cyberplace.social I especially like that there is a 45GW power shortfall for all the new datacentres in the US, and no one seems to know where that's going to come from. For context, that's about the size of the whole of the UK's generation.
@steve@mastodon.nexusuk.org @GossiTheDog@cyberplace.social According to the article, that's 10% of the entirety of the United States. And that's spread out across a lot of land. Focusing that much in a few data centers is virtually impossible and this will have a very very bad effect on power grids.
@nazokiyoubinbou@urusai.social @steve@mastodon.nexusuk.org @GossiTheDog@cyberplace.social
That's why they're building/reactivating nuclear power plants. Nuclear is much more expensive than renewables, but if you need a gigawatt right here, in this location, nuclear is almost the only solution (to a problem we shouldn't have).
@AdmSnackbar@mastodon.social @steve@mastodon.nexusuk.org @GossiTheDog@cyberplace.social And that is not going to end well.
(Not because I consider nuclear energy to be the horrible thing people see it as, but I sure don't think building plants just to power LLMs and data centers is the answer and, more importantly, these are not the people who should be in charge of such things either... You know they're going to skimp in every way possible and nuclear fission is a thing you do not skimp on...)
@nazokiyoubinbou@urusai.social @AdmSnackbar@mastodon.social @GossiTheDog@cyberplace.social "move fast and break things" when dealing with nuclear reactors π¬