Will the bubble burst

[pcprogrammer]

Just watched this video and do wonder if it is right.

The hype is being fulled for sure and driving up prices of computer stuff without a doubt. In September 2025 I bought two new computers with reasonable specifications for about 800 euros a pop. My brother is looking into buying one with the same specifications and comes to a total of about 1450 euros. DDR5 memory has more than doubled in price, basically because of the AI bubbles need for memory.

[iMo]

There are megatons of YT videos showing how we are in the AI bubble and why and how it will burst.
The large investors know well about the potential bubble so they have adjusted their strategies already.
It could easily happen the bubble will not burst in the way they describe in the YT videos, but the AI situation will continue for many years to come (my bet a decade) and slowly becomes an integral part of the world's business cycles..
PS: the ddr5 mems issue might be solved by new manufacturers in China, for example.. They already do 5%..

[pcprogrammer]

There are megatons of YT videos showing how we are in the AI bubble and why and how it will burst.

I was offered just the one by the Youtube "AI" 

The large investors know well about the potential bubble so they have adjusted their strategies already.
It could easily happen the bubble will not burst in the way they describe in the YT videos, but the AI situation will continue for many years to come (my bet a decade) and slowly becomes an integral part of the world's business cycles..

We just have to wait and see.

PS: the ddr5 mems issue might be solved by new manufacturers in China, for example.. They already do 5%..

Depends on the quality of the chips they provide, but not a bad thing that there is upcoming competition for the mainstream memory manufacturers.

[SiliconWizard]

The issue will not be solved by any *reasonable* amount of competition.

If China increases its RAM production and western companies start using that, prices will skyrocket again in no time, most of the production will still be diverted to "AI", and back to square one within probably just a few months. The only difference here may be that China will probably refuse selling more than a relatively small percentage of its production to foreign companies, to secure their own market.

I see no solution other than either an AI crash or strict regulations coming into place, preventing any single industry / customer from grabbing more than a certain percentage of the overall semiconductor production. This kind of regulations could even be considered national security measures.

[free_electron]

they are only inflating the AI balloon. Just like when you blow up a balloon the initial push to start inflating is hard. Once it has expanded a bit it becomes easier and the balloon expands faster. When you approach the end of inflation the balloon expands slower.
AI is now past the initial inflation point ( balloon analogy) and we are in rapid expansion mode.

Ai is far more than ChatGPT and the likes. The field is rapidly expanding into transportation and robotics and more.
Linking an AI with real-world interaction is where the future is. There are many real-world tasks where a system needs to be able to react to the environment. A fixed-program system can't do that. You cannot write a program that can accommodate for all possible variations. Such a decision tree would be enormous and take an immense amount of time to define and code.

That's where AI comes into play. It can either help you create the decision tree , or it can replace the decision tree.

[Zucca]

I hope, so I can upgrade my RAM and NMVE without opening a loan.

[pcprogrammer]

That's where AI comes into play. It can either help you create the decision tree , or it can replace the decision tree.

I can see lots of fields where a neural network setup can bring a lot of benefits. Medical research or diagnostics are fields that can greatly benefit from it.

The hype around the likes of ChatGPT in the consumer world is kind of idiotic. It is seen as the big oracle by many, and that can lead to problems, like the child porn stuff scandal of Grok, or the suicide shit from ChatGPT, or the somewhat false idea that these AIs can write the code for you to make your Arduino project work without a hitch.

Just like humans it takes proper training to make it do things properly, and these grab the info from the world wide web AIs are, in my opinion, not properly trained and spit out a lot of garbage.

Something the general population seems to overlook. 

[Analog Kid]

That's where AI comes into play. It can either help you create the decision tree , or it can replace the decision tree.

I can see lots of fields where a neural network setup can bring a lot of benefits. Medical research or diagnostics are fields that can greatly benefit from it.

The hype around the likes of ChatGPT in the consumer world is kind of idiotic. It is seen as the big oracle by many, and that can lead to problems, like the child porn stuff scandal of Grok, or the suicide shit from ChatGPT, or the somewhat false idea that these AIs can write the code for you to make your Arduino project work without a hitch.

Just like humans it takes proper training to make it do things properly, and these grab the info from the world wide web AIs are, in my opinion, not properly trained and spit out a lot of garbage.

Something the general population seems to overlook. 

I think we can be charitable here and give the "general population" a break. After all, they weren't the ones who demanded that "AI" secret sauce be added to everything having to do with computers: that was our Eeevil Overlords (and the usual greedy gold-diggers) who are pushing it on us.

Hopefully they (the great unwashed masses) are slowly waking up to what a tremendous load of horseshit the whole "AI! AI!" thing is. (This is already happening in places like YooToob videos where "AI"-generated slop is being called out by commenters.)

Disclaimer: Not that there aren't applications where "AI" can be useful. But not in its current overblown, over-hyped state.

[M0HZH]

The AI industry is exposed in a way that has never been seen before, investments are absolutely insane compared to the (current or potential) returns. The only company with a real advantage is Nvidia, the larger players are diversified so they're relatively safe, OpenAI and their investors will probably take the massive hit.

The technology already hit the plateau. The play is now to put AI between humans and machines (agentic AI) so things become much more efficient, but it's hard to yet see a scenario where that pays back at the large scale. Next step is self-evolving AI (aka "AGI"), still not sure what we can do with it besides military / scientific research. It will take another 1-2 years before the new datacenters properly enter production and by then there needs to be a way to properly monetize it; best case is it fizzles out in 5 years, worst case market panics and there's a massive crash in 2.

[negativ3]

Energy is going to be the bottleneck.
Especially in areas who currently embrace the net-zero fallacy.

[iMo]

The issue will not be solved by any *reasonable* amount of competition..

The gamersnexus vid shows the ddr5 production consumed by AI tycoons today is 900.000 wafers per month..
The CXMT made 270.000 wafers per month in 2025 (from the vid). Let us assume 300mm wafers.
So it is rather high percentage even the yield per wafer might be lower.
On the other hand CXMT will not be selling them for the mid2025 prices, but say 15-20% lower than the three other manufacturers (when the ddr params are compatible).
IMHO the prices of the ddram5+ will never come back to mid2025 prices..
Edit: the prices of the ddram5+ and ssd disks..

[paulca]

The thing is.  The dot.com bubble burst, but it didn't stop the eCommerce revolution.  It too was caused by "early hype" for a future that didn't exist yet, until it did.  The whole "Dark fibre" thing etc.

Early hype for AI is off the charts, so people are getting over invested.  It's standard procedure.  The large and the smart will pull out at the last minute and cause the crash.  The suckers left behind get fleeced.  They write their debts off between each other and regroup for "bubble 2" inflating.

People are duldrumin around with LLMs and text bots, in 1 dimensional AI transformers (dialup internet) while the future is in multi-dimensional AI transformers and many other things.  There just isn't the compute power (or possible power itself) it run these "future" ideas yet, but that time will come.

"Yes we know it sort of sucks right now, but trust out vision." is that they are saying.  They are thinking about "Fibre to the home" while we call, "All hype" from a dialup internet connection - Dot.Com all over again.

[John B]

The thing that really blackpills me is the feeling that the people who have invested absurd amount of money into hardware procurement, and screwed the market at large i the process, won't take the fall if the bubble pops.

[Stray Electron]

Energy is going to be the bottleneck.
Especially in areas who currently embrace the net-zero fallacy.

   Not just energy but also water.  Many of the locations that they want to put AI data centers in such as Arizona, Texas, California and Florida already have real water shortages or have border line shortages. Right now, the Florida legislature is considering a law to prevent AI data centers from getting highly preferential energy rates that would push the real cost to the regular consumers. And they are considering a bill to prevent data centers from using excess amounts of water (needed for cooling). Some of the power plants being proposed in order to power the AI data centers will also require massive amounts of water.

[pcprogrammer]

They should have asked AI first to provide a solution for the energy and water consumption problem. 

[coppercone2]

the interesting thing is that bubbles are not like something that the most recent generation has seen. You can say that housing is a bubble but its more abstract I think, it was more of a general fuckup with management policy then a technology failure. Alot of the AI thumpers were born after the dot com crash! Naturally someone born after something will take less from it, because it seems like ancient history, plus the 2008 bubble had alot more effort put on it to manage the fallout, because things were already pretty messed up from the dot.com bubble and it was basically war-time (big wars make the government bail things out which should die, because it feels like you might lose if you don't, they were thinking national security instead of fair/healthy economy, given how much fear there was, first big attack on the USA since pearl harbor). I have a feeling the govt is thinking "we can't do a big bail out again", that is like those helper options you get on who wants to be a millionaire, you get to use one every 60 years or so... the problems they are dealing with now were more manageable and distant and there was more optimism that 'meddling' could be 'compensated for'.

Older people can remember those instances where they heard someone speak about dot.com bubble things and it seemed like the hottest thing in the world and you wished you were in on it, and then it just vanished without a trace, and its like eerie. 20 somethings don't have that uncomfortable memory (which gave you direct context, that you were there, you measured the 'temperature' of the situation, you remember what people in real life were saying during that time, what it felt like, what excuses the wiseasses were giving, the 'vibe' and so on) and reading about it is just data to them that can't be gauged well. Renembering how optimistic and futuristic it felt.. and then see it implode..

The scary thing is that the risk-evaluation of AI related investments.. might be going by how the government feels about 'physical property' (housing related) immediately following a 'surprise attack' that lead to war after a long peace (the unprecedented 2008 response after a decade of 'prosperity' following the fall of USSR until WTC), instead of what is likely to happen as a response to the next bubble (which might be the old 'hit the road jack'). It does NOT look like the federal government is looking to save face, make friends, etc right now. I expect a hardliner response, since we are seeing now that they will mobilize federal authorities to deal with things instead of trying to placate civilians. Basically I think they are liable to 'belt' right now instead of time outs, and no one is going to go out of their way to prevent a few 'detroits' from forming when there is a problem. I hope they figure out how to de-escalate this bubble privately in a intelligent and reasonable way, because someone (mr orange?) is gonna say FU when they try to pass the buck up too high and everyone will suffer.

Also the 2008 bubble was like high level and confusing to understand, it was unclear what exactly was going wrong, that "bankers were doing something bad" I guess was the thought the majority of people had, but it was like  unclear and it seemed like something that could be handled internally (i.e. look at occupy wall street protests (which I think were related to 2008), the failure of it was that no one knew what the hell was going on and what they should be angry at, so the general public was kinda ambivalent, or at least it was confusing to pick a exact target. I feel like from 2001 to maybe 2009... it felt like a portion of the US government was on crack making alot of panic decisions that don't make too much sense.

And I don't think that COVID related bail-outs are anything to go by either, because everyone was scared that they will get killed by nanoscopic little floaty mines, at least for a while, I think they had the 'alien invader' mentality

Additionally, with AI, the general public could very easily and clearly form the thought that "these guys are trying to replace us all with robots", so the public has like a seemingly rational and solid reason to just.. get rid of it and think they are doing the right thing, its like almost a no-brainer, compared to something with financial flow charts. I mean when they are going to be incredibly easy to rile up to fanatic levels because of this line of thinking, and that means that you can get away with a very.. heavy  handed... response with alot of support, even if its detrimental to the supporters financial well being. Not saying I agree with this thought, I just think that it will be a very easy angle to play in the future when there are problems.

What I am getting is that people seem to have the idea that things will quickly be OK if this bubble bursts (i.e. the sentiment i see about "so what if dot com burst, we have internet now" (I wonder if this is why the internet like sucks though)), but I expect the consequences would be alot worse then they are hoping for, and this itself fuels the bubble, gives it more integrity (because the consequences are not thought to be severe enough, so the normal 'pop this thing now' response is dampened)... and it could be a real blow out.. with longer recovery times and more dire consequences then people think.

I wonder if the so called 'dark ages' could be thought of as a post-bubble (roman influence bubble). That is another interesting thought, the more bubbles you survive, the more risk you decide to take..

[ejeffrey]

   Not just energy but also water.  Many of the locations that they want to put AI data centers in such as Arizona, Texas, California and Florida already have real water shortages or have border line shortages.

The water usage is not a requirement. Datacenters use water because it's the cheapest way to get a lot of cooling.  Plenty of data centers used closed loop cooling and don't consume a drop of water other than the small amount needed to initially fill their coolant loops.

The problem is that water is a negligible fraction of a data center operating cost.  You basically can't raise the price enough to  make it non economical.  If you tried you would drastically impact people who need water for necessities like drinking and washing.

If you just ban using evaporative cooling in data centers as several places are considering they will build closed loop coolers.  That will basically solve the problem at least for water. 

[ejeffrey]

Alot of the AI thumpers were born after the dot com crash! Naturally someone born after something will take less from it, because it seems like ancient history,

Nah, those are mostly grunts.  The people at the top really pushing this: your Sam Altmans, your Marc Andresans, and your Sergey Brins all experienced the dot com bubble.

[coppercone2]

I think the youngers are more obsessed and devoted to it, some what because of theories of brain plasticity, they don't see it as a threat because learning is still easy to them (personally I think its just that you slowly get lazy and you need to work pretty hard to get back into 'learning mode' after some advanced age of 2something, and its like not cool with other adults and it means you go back to being a nerd, so no one really wants to do it, college is like a social excuse that you get a 'limited time nerd permit' but you are still cool afterwards ).

The older CEO and upper management have enough golden parachutes and such that its not a threat to them, and younger people feel like they can adapt (i.e. the world feels open still), and alot of the power of AI seems to be geared at amusing them (i.e. silly picture generators).... I think its like circus&entertainment for the 'base' of the power. All those leaders you mentioned would probobly laugh their way through the great depression too....

After all generals don't 'execute' war, the grunts do. The thing I noticed is that the grunts might be particularly oblivious and mis-informed this time around. I think the.. precedents they have for risk management right now (and the projected 'consequences') are wrong and its leading to unusually risky behavior. And part of it is cultural in the sense that the internet makes you feel more informed but I think that was only for a small time period, right now its so covered with waste and corporate that it might actually just make you overly confident when making 'projections' that help dictate your behavior.. in the sense that people would have to be spending way more time fact checking, verifying, consulting and searching then they do to get the actual level of confidence they think they have.

Makes me think about when at work sometimes there is some technical BS floating around that maybe there is something there that could be doing something, and you get some opinions and 'facts' from people.. and then when you hit the books you figure out by the first equation that its a bunch of rotten information that was stuck in a telephone game for a long time and the soft measures and stuff that were being taken (based on cursory 0 effort internet advice) are like.. insanely wrong/useless and based on a misunderstanding of the underlying issue

[Simmed]

https://www.windowscentral.com/artificial-intelligence/microsoft-confirms-plan-to-ditch-openai-as-the-chatgpt-firm-continues-to-beg-big-tech-for-cash

its starting ?

Google Deepmind co-founder and Microsoft AI lead Mustafa Suleyman suggests that the big tech firm is moving away from OpenAI reliance, as the latter's financials look increasingly dire.

In a move that is perhaps less surprising than you might think, Microsoft seems to be readying up to dump OpenAI.

Right now, Microsoft's entire AI operation is powered by ChatGPT and other OpenAI models, including DALLE 3. Microsoft has shown impressive growth and demand for its enterprise-grade AI tools, including things like Microsoft 365 Copilot and Github Copilot, even if its consumer-level efforts have largely fallen flat.

Microsoft and OpenAI are long-rumored to have endured something of a tumultuous relationship. Microsoft was a very early investor in OpenAI, and won itself some incredibly lucrative contracts as a result, including some forms of exclusivity over OpenAI's models. Indeed, Microsoft still holds 27% of the new "for profit" arm of OpenAI, and maintains IP rights for OpenAI's models until 2032.

However, Microsoft and OpenAI re-worked aspects of the deal last October, freeing up OpenAI to seek compute from competing cloud firms, and allowing Microsoft to divest itself of some of the endlessly spiralling risk.

OpenAI is notoriously on the hook for over a trillion dollars in future compute spend contracts, with big tech companies like Microsoft, Amazon, Softbank, and others artificially propping up the company. Run by Sam Altman, OpenAI has been mired in almost constant controversies, its balance sheet notwithstanding. The firm has so far made zero dollars, and requires near-constant cash injections to stay afloat.

Things may be about to get even rockier for OpenAI, as Microsoft AI chief Mustafa Suleyman just confirmed to FT (paywalled) that the firm is gearing up to ditch OpenAI's models.

[Cyclotron]

AI isn't going anywhere. It's here to stay, its been here a long time in a lot of ways. Today's AI hot topic are the LLMs.

The possible bubble is not weather AI hangs around or expands. It will do both.

The bubble is that investors pour billions into over building and poor business models. Both are the same issues the "DotCom" bubble suffered.
There's some risk here but not nearly like the dotcom era. The biggest impacts will likely be discoveries that allow for faster and more efficient models. But the appetite for work loads on LLMs is so huge at the moment that optimizations aren't likely to impact build rates negatively.

The only part that I disagree with some of my peers about is the impact to employment. I believe AI technology will cause significant short term and long term impact to the job market for many fields. A good number of my colleges feel that this isn't true and that people will simply be more productive. My example is that the refrigerator didn't make the ice delivery man more productive. I think they are short sighted on the application of AI on various jobs.

We'll see....

[sleemanj]

I'm not sure it is a bubble, Ithink in the same way as we were taught about the industial revolution in our schooling, generations in the future will learn about the AI revolution.

[screwbreaker]

The bubble can't burst.
It's now understood that AI can generate any content and make it seem real.
Every government, or at least none of the major governments, will want to leave this technology in the hands of another nation.
The problem is that to achieve short-term progress, the only way forward is to push resources hard. More RAM, more computing power, more electricity, more money.
In the short to medium term, I see it as difficult to recover the money by selling AI-based applications and services.
For now, the AI ​​bubble will be kept afloat by governments, then we'll see.

[Flogiston_Wave]

PS: the ddr5 mems issue might be solved by new manufacturers in China, for example.. They already do 5%..

Unlikely.
Chinese rare not stupid.
Why would they sacrifice themselves to help their adversary that does everything he can to sink them ?

Their RAM production will be prioritized  to benefit them.

If we get to see Chinese DDR5, it will be their substandard stuff that they get to sell while developing better models.
And prioritize HBM.
And even that is questionable at any appreciable scale.

[Cyclotron]

PS: the ddr5 mems issue might be solved by new manufacturers in China, for example.. They already do 5%..

Unlikely.
Chinese rare not stupid.
Why would they sacrifice themselves to help their adversary that does everything he can to sink them ?

Their RAM production will be prioritized  to benefit them.

If we get to see Chinese DDR5, it will be their substandard stuff that they get to sell while developing better models.
And prioritize HBM.
And even that is questionable at any appreciable scale.

China has as many adversaries as it makes. Ruling communist party in China only does things to benefit that class alone at the cost of its own people, every other nation's people and so on. If it takes social credit score to keep people in line, they do it. If it makes the ruling class more powerful and is to their benefit they will sell whatever to whomever.

Count down begins to the "What aboutism country "Fill in the blank".....

There's a myriad of countries that can setup and run fab shops at near the same cost as it is in China, maybe cheaper. China's cost controls have been either the extreme low pay of factory workers which has no bearing here or the subsidization of cost by the government.

I predict, if the folks doin the math show there's money to be made, lots of places will spin up memory fabs. 10 to 20 or so nanometer fabs are perfectly suitable for memory fab. These are not cutting edge fabs by any definition.

Edit:
I forgot to dd the URL to the study about the ruling party being for itself.
https://sccei.fsi.stanford.edu/china-briefs/how-much-does-party-membership-pay-measuring-impact-communist-party-membership-wealth