Is this the Beginning of the End for the AI Bubble?

[K3Nv4]

not that part of it. no coal burning in the area I was around.

Nothing is stopping you from breathing in oil toxins but I'd rather have the clean air and deal with data centers and watch oil companies reduce in size.

 

[MPRTwice]

Nothing is stopping you from breathing in oil toxins but I'd rather have the clean air and deal with data centers and watch oil companies reduce in size.

actually, my lungs were having a much easier time by oil than by coal. I almost forgot what it is like to not have double vision/consistent silent chest asthma. Many report similiar, even air force pilots who do bio-conditioning for high G-Force environments.

If an air force pilot can't live here without allergy meds then you should know it's bad, but their visits to Texas? They were able to breathe well.

coal is factually much more dangerous to breathing than being around oil in general. I am also living proof of that.

 

[Vixi]

Just some extra clarification on that number.

Out of 5,300 about 90% of those are traditional data centers. Those use 5-10x less energy than AI centers. The remaining are the energy and water guzzling AI centers along with most new centers which are being built today.

Traditional datacenters are still guzzlers and completely unnecessary.

 

[Nakamichi]

Traditional datacenters are still guzzlers and completely unnecessary.

Arent traditional datacenters basically hosting gbatemp?
i cant see a lot of bigger pages actually selfhosting

 

[lightwo]

Traditional datacenters are still guzzlers and completely unnecessary.

Cloud does often cost more to use than having own infrastructure. It is also more marketable as it's a single service that covers all of processing, uptime and almost everything else. But in terms of datacentres, companies need their own infrastructure, and just because they use or don't use cloud doesn't mean there isn't a computer running their services somewhere... so the only alternative in regards to resource use would be to not have computers at all.

Of course, I'm not talking about "AI"/LLM training/hosting datacentres. The society would be better off without them ever again.

 

[titan_tim]

Traditional datacenters are still guzzlers and completely unnecessary.

Depends on if you think the Internet is a necessity I guess. Companies can have their own racks in their offices, but in the end it's a nightmare for them to maintain and keep clean. In the end, keeping servers centralized like that helps to keep cooling everything a little more energy efficient than if they were spread out all over.

Post automatically merged: Jun 12, 2026

A fun quote that I saw earlier which is pretty damn true:

The reason RAM prices are so damn high is that a massive amount of not-yet-manufactured memory was bought with money that theoretically exists to be put into GPUs that haven't been made, to be put in data centers that haven't been built, powered by infrastructure that's not built yet, all in pursuit of profits that may never come.

 

[Sir Tortoise]

Generative AI has been massively subsidized in the hopes that, if they just keep piling more money and data into the furnace, it will become essential to daily life and then they can raise the price and make a profit.

Unfortunately, while LLMs do some interesting stuff, there just isn't that much demand for technology that sometimes produces half-decent results and sometimes shoots you in the face.

It might be rising user costs that pop the bubble, or it might just be companies failing to repay the trillions in debt they've racked up to build these datacentres. The AI buildout is literally consuming the vast majority of the entire world's investment funding, and all we'll have at the end of it is some extremely niche and expensive datacentres, full of GPUs that can't be used for anything else and burn out after a few years. And to run them, we only need to sell another 10 billion ChatGPT subscriptions.

That is assuming the datacentres even get built. Right now they are facing a few minor problems with reality, like, we're acting like this is a videogame where you can just build 500 new nuclear power stations if you have the imaginary money for it. But we still have to pretend like the datacentres are actually being built, because the stock market runs on Looney Tunes physics and if we look down we all die in an economic crash that will require the Great Depression to be renamed like the Great War was.

 

[Taleweaver]

Depends on if you think the Internet is a necessity I guess. Companies can have their own racks in their offices, but in the end it's a nightmare for them to maintain and keep clean. In the end, keeping servers centralized like that helps to keep cooling everything a little more energy efficient than if they were spread out all over.

Seconded. We moved our company's servers a week ago. One suffered a hard drive failure (or rather: two, as it was a RAID5 setup), resulting in the loss of a program that should've been retired over a year ago (but it's still used because people). Suddenly it was as if that one program was the centerpiece of the company.
Yes, there are failures on data centers as well. But there is maintained by people who do it professionally and maintain it as their day job rather than it being one of the hundreds of other things. Fixes happen in matters of hours, not days.

But more on topic: the most disturbing thing I've read is that 99% of ceo's would replace employees with AI if it were economically viable. I can't say i fear for my own job (I'm more the one who'll fix its hallucinations), but it highlights how little ceo's as a whole understand economy. Does it matter if you can make more output with less people if the market also shrinks because of it? Aka: who will buy stuff when AI took everyone's jobs?

 

[Reiten]

Regarding the costs, for an employee the cost is pretty much know, it's their salary, for AI not so much. There is the implementation cost, one time and mostly known, but even here additional costs can creep. Next are the monthly costs, depending on how stuff gets implemented, there could be maintenance and API costs, and a bunch of others(electricity, server rent, etc.). Most analyses I've seen, show that in the end AI costs the company more then the employee they replace without any real boost to performance. A recent study actually suggests that companies that give AI as a tool for their employees see an increase in performance, but most of the ones that replace employees with AI see a decrees in performance.

Fun story an admin friend told me, a company replaced on of their accountants with AI, the job was to mostly pull data from various sources and compile financial reports. Another of the accountants used these reports for IRS(well the equivalent over here) related documents. One day the AI seems to not have been able to get data from some of the sources, but instead of informing someone, it decided to just hallucinate up numbers, which ended up being submitted to the IRS and got the company a fine.

 

[DuoForce]

There's no such thing as an "AI Bubble" AI is here to stay. People in the 80s/90s thought the internet was also just a fad/trend and it merged into our daily lives. The same will happen with AI. If there was a bubble it would've popped by now. Get ready for a decade of "guys the AI bubble is gonna pop fr this time!" over and over again

 

[Taleweaver]

Regarding the costs, for an employee the cost is pretty much know, it's their salary, for AI not so much. There is the implementation cost, one time and mostly known, but even here additional costs can creep. Next are the monthly costs, depending on how stuff gets implemented, there could be maintenance and API costs, and a bunch of others(electricity, server rent, etc.). Most analyses I've seen, show that in the end AI costs the company more then the employee they replace without any real boost to performance. A recent study actually suggests that companies that give AI as a tool for their employees see an increase in performance, but most of the ones that replace employees with AI see a decrees in performance.

Fun story an admin friend told me, a company replaced on of their accountants with AI, the job was to mostly pull data from various sources and compile financial reports. Another of the accountants used these reports for IRS(well the equivalent over here) related documents. One day the AI seems to not have been able to get data from some of the sources, but instead of informing someone, it decided to just hallucinate up numbers, which ended up being submitted to the IRS and got the company a fine.

I'm not sure i agree on the cost, but it's regarding perspective. For AI, the monetary cost is lower than an employee. But that's hardly an achievement, as the jobs that require no creative input have mostly been automated for years.

It's the cost of training that's a major hurdle, and I presume a lot of soepje that's hard to quantify but very essential.
Take your example: people tend to believe output by a computer if it strokes with what they want to hear. It doesn't take a genius to take well formatted input and output it in another program. But reality is often messy, like when a source changes url, some vendor decides to change the output or there's an outage somewhere. I won't say that people don't make stuff up in such instance (the cynic in me even presume that ai has learned hallucinations partially from humans, in fact), but it often raises flags or checks to make sure that the situation gets resolved. AI isn't accountable, and it has to be trained on a specific job.

But there's a worse factor: I've read (and my experience matches that) that knowledge isn't the de facto bottleneck. That is: the major screwups I've seen weren't done by workers but by managers by insufficient checks of what projects had on others. It taught me that managers are better at explaining and convincing than knowing what they have to do. The people under them tend to fill them in... AI won't. That won't be normally calculated in cost, but it's certainly a factor.

Post automatically merged: Jun 13, 2026

There's no such thing as an "AI Bubble" AI is here to stay. People in the 80s/90s thought the internet was also just a fad/trend and it merged into our daily lives. The same will happen with AI. If there was a bubble it would've popped by now. Get ready for a decade of "guys the AI bubble is gonna pop fr this time!" over and over again

LOL...no. I knew internet was gonna become mainstream the moment it stopped clocking up the landlines. But that doesn't mean there wasn't a bubble... Just look up 'dot com bomb'. It just meant the end of the idea that everyone with a website would become a millionaire.

Same with AI: nobody says there won't be a chatbot anymore in the future. But right now there's an arms race to reach... Well, AI singularity, really. But it won't mean that they'll all be profitable. In fact, given the huge costs (way more than the common internet development), you can bet that there'll be massive losses.

 

[Axido]

There's no such thing as an "AI Bubble" AI is here to stay. People in the 80s/90s thought the internet was also just a fad/trend and it merged into our daily lives. The same will happen with AI. If there was a bubble it would've popped by now. Get ready for a decade of "guys the AI bubble is gonna pop fr this time!" over and over again

Ever heard of the Internet bubble? That thing did in fact pop back in March 2000. The Internet itself survived that. It's not about whether the technology stays or not, but how much it's overvalued at a certain point in time.

AI isn't where companies like openAI, Google, Meta and Anthropic want us to believe it is. It will take an unspecifiable amount of time until there will be something that can actually be called artificial intelligence. The thing currently called AI is bound to fail as it's just an unfinished expensive bullshit machine that not even its creators understand.

 

[DiscostewSM]

Ever heard of the Internet bubble? That thing did in fact pop back in March 2000. The Internet itself survived that. It's not about whether the technology stays or not, but how much it's overvalued at a certain point in time.

AI isn't where companies like openAI, Google, Meta and Anthropic want us to believe it is. It will take an unspecifiable amount of time until there will be something that can actually be called artificial intelligence. The thing currently called AI is bound to fail as it's just an unfinished expensive bullshit machine that not even its creators understand.

Yeah, that's what I was thinking. These bubbles did happen, and they did pop. That doesn't mean they won't exist afterwards. It just means there won't be as much of a push to have it for the sake of it. Right now, "AI" is far more damaging to civilization than the internet ever was, so that bubble may burst to a much greater degree.

 

[Nakamichi]

Ever heard of the Internet bubble? That thing did in fact pop back in March 2000. The Internet itself survived that. It's not about whether the technology stays or not, but how much it's overvalued at a certain point in time.

AI isn't where companies like openAI, Google, Meta and Anthropic want us to believe it is. It will take an unspecifiable amount of time until there will be something that can actually be called artificial intelligence. The thing currently called AI is bound to fail as it's just an unfinished expensive bullshit machine that not even its creators understand.

i think over-valued describes it well. Whenever a Tech-CEO opens their mouth nowadays all they ever talk about is AI, and every single time they do they sound like they've completely lost their mind.

 

[DiscostewSM]

Here's the thing. I don't want AI to go away. I don't think any of us really want that. But, what I do want from AI is to be meaningful. Useful. Not just a toy for the billionaires to play with at the expense of the people. One of the things I want to see AI be used for is medical research. Science.

 

[Nakamichi]

Here's the thing. I don't want AI to go away. I don't think any of us really want that. But, what I do want from AI is to be meaningful. Useful. Not just a toy for the billionaires to play with at the expense of the people. One of the things I want to see AI be used for is medical research. Science.

i like local AI, its fun and means i never have to write another letter myself ever again, i just dont like Cloud AI. I think its a very bad idea for absolutely everyone.
And id love to see Jensen eat his words and fall on his fat, stupid mouth.

 

[fst312]

I don’t mind ai at times but I feel like it’s going to be one of those other things that take peoples jobs away.

 

[Sir Tortoise]

There's no such thing as an "AI Bubble" AI is here to stay. People in the 80s/90s thought the internet was also just a fad/trend and it merged into our daily lives. The same will happen with AI. If there was a bubble it would've popped by now. Get ready for a decade of "guys the AI bubble is gonna pop fr this time!" over and over again

The situation is inherently unsustainable, something like 50% of the stock market is reliant on a technology that literally burns money. Not a single AI company makes money besides Nvidia selling shovels.

www.isaiprofitable.com is a nice visualisation of the scale of the problem

Either they magically achieve world-changing technology like sentience (no) or it bursts, we can't continue as we are forever. The chatbots are cool and all but they're not trillions of dollars cool.

 

[DiscostewSM]

The situation is inherently unsustainable, something like 50% of the stock market is reliant on a technology that literally burns money. Not a single AI company makes money besides Nvidia selling shovels.

www.isaiprofitable.com is a nice visualisation of the scale of the problem

Either they magically achieve world-changing technology like sentience (no) or it bursts, we can't continue as we are forever. The chatbots are cool and all but they're not trillions of dollars cool.

It's very much a gold rush scenario. It's not the people trying to get gold that are getting rich. It's the people supplying them with resources, tools, etc that are raking it in.