So more or less this: https://youtu.be/T9d6lrvzM9o
I long for the headline that Sam Altman in total collapse ate his own pistol.
once thiel stops pulling his string.
You won’t get the GPUs. Most of the AI hardware is not usable to you. They didn’t just sell all the desktop RAM and GPUs to AI companies, they dedicated most of the production to data center hardware that you can’t use. Relief isn’t coming, what you need doesn’t exist.
I’m sure with all these AI startups there’s plenty that used consumer grade stuff because that’s what they could get
Either way though, I run enterprise servers in my homelab so I stand to benefit even if mostly enterprise server RAM and data center class GPUs hit the market
You can rent consumer grade GPUs on cloud servers for your AI computing, so yes
And they’re not going to switch production back to consumer hardware either. Those data centers will be used to sell virtual desktop subscriptions.
The hardware in those DCs isn’t suitable for virtual desktops. It’s specifically designed for doing lossy matrix multiplication, something that’s only useful for a certain type of “AI” tasks.
At that point I’m willing to raid a data center
Why the fuck would anyone buy this. Come on.
Because they will have no other option once their existing system dies. In this scenario actual PCs will be something only the wealthy can afford. Everyone else gets dumb terminals.
And honestly, VDI is great from a general user standpoint. It’s the companies who run it that make it a bad option.
Well, they’re also a bad option when the data centers hosting them get bombed
I’m far from an expert on these things, but I have a tough time believing that latency won’t be a critical issue for many consumer applications.
With ftth, latency isn’t as big of an issue as it used to be.
Because people are fucking idiots.
Federal Reserve hasn’t raised the interbank lending rate yet. So speculation over values 60 year domination schemes over actual consumer value vs cost now.
As long as you give banks artificially cheap loans with money printed on demand, the bubble can keep going and they can fund R&D we will regret later. Regret it because it doesn’t need to satisfy a customer or use only a reasonable amount of resources. It just has to attract investors with the promise that it will attract investors with the promise that it will attract investors. It doesn’t need a customer when you’ve messed with interest rates.
Your econ class only told you lowering interest rates increases jobs in the short term. It didn’t tell you why. It adds jobs doing less strictly productive things. Long term all economies tend to the same employment level anyway. And long term lowered interest rates tend a market toward bullshit. We get more work but less consumer satisfaction. That’s really to say the cost of living goes up with respect to an hour of work. And you have to compete as a consumer for resources with firms that have access to infinite cheap money.
The bubble popping probably won’t help since they’ve stopped making consumer hardware altogether.
That’ll start to turn around in a few months once manufacturers lose that easy AI money.
Even if the AI bubble collapses tomorrow (it won’t), to see lower consumer prices you would either need old stock to be marked down or the fabs that are currently making AI cards to switch to making gaming cards and those cards to be manufactured and come through the market. That will take months. Not only does the bubble need to pop, but everyone involved needs to believe and accept that it has popped, otherwise they will continue to keep prices high and AI production high in the hope that their plans remain correct.
Remember that it’s psychologically much harder to accept making a less money than you thought, which happens when you mark down prices, compared to making more money than you thought, which happens when you mark up prices! This is why prices tend to be “downwards sticky” and so, instead of going down when demand drops/supply improves, they instead tend to stagnate while other prices continue to rise gradually with inflation.
That will take months.
Dude, I’ve been waiting to buy a new card for something like a decade. I can wait months.
but everyone involved needs to believe and accept that it has popped
That’s the very definition of a bubble popping. Something’s a bubble when people believe it has value when it really doesn’t. The bubble popping is the moment everybody realizes the value was an illusion and there’s a rush to get out. As long as people still believe that there’s still value, it’s still a bubble and hasn’t popped.
We’re already in the stage of the bubble when many people know it’s a bubble and are doing everything possible to hedge against the bubble popping, while others just keep pouring more money in.
I think someone saying, “come on, burst already” does not see the bubble popping as “when diminishing confidence in AI has reached all bits of the economy” because they see it as something likely to happen soon, and that ain’t happening soon. However, we may see reduced AI investment from some quarters soon.
It’s also worth recalling that stock market bubbles can and do pop very suddenly - I don’t think it’s that reasonable to take a definition that spreads it out over months, but maybe that’s what you’re saying.
What I’m saying is that when it happens it will probably be fast, and it could happen anytime. I personally think it still has at least half a year to go because I think the tech / AI companies are so close to the Trump administration that it will jump in to save them if there’s a problem, which will delay the pop by a few months.
But, I can’t see the bubble deflating slowly. The bubble is just too big for that. They’re betting on AI Jesus. Investors won’t be satisfied with a normal technology which is somewhat useful, not after pouring trillions into it. Just to break even, the AI companies need to reach revenues of multiple trillions of dollars. That’s almost 10x more than Google makes, and it’s a mature 20-year-old company that has a monopolies in multiple areas.
Just to break even, it has to be something that every single person on earth spends hundreds of dollars on each year. And, since that includes multiple billions of people who don’t have hundreds of dollars to spend, it’s really multiple thousands of dollars spent by every person, or spent on behalf of a person in a developed country.
Also, when the AI bubble pops, nobody will have any GPU money because they need it to bail out the AI companies to prevent the global economy from blowing up and flinging the entire cosmos into Satan’s buttcrack.
By that time, nobody will be able to blink without AI because everyone is using it all the time for everything even though everyone knows the AI bubble is going to pop and the world will go to shit because everyone depends on AI.
I have no idea what we’re doing, but I’m sure it’s fun to watch from far away.
Who said anything about wanting gaming GPUs?
ah I didn’t realise you actually wanted to release your own model 🤔
Release? Probably not
But the LLM tech is cool to experiment with, and there are useful use cases, but trying to run frontier class open source LLMs takes a boat load of VRAM lolol
Oh no, months. That makes solving a problem implausible if the benefits can’t come till months.
I don’t wanna rent the hardware for my instance forever :(
Remember how the housing bubble never popped?
Uh are you ignoring the huge financial crises, and how banks had to be bailed out?
I remember housing costs NEVER going down.
https://fred.stlouisfed.org/series/ASPUS
2005-2011.
The housing costs never reached some arbitrary “fair” value, but they most definitely did go down.
Yeah… :(
How long is this going* to take?
@cm0002 No good answer for RAM right now but with respect to GPUs, consider Intel Arc. I purchased a B50 Pro for a whole of $300, it’s turned out to be a reasonably solid performer. I have SR-IOV working on it now a vf for a windows vm, and one for LLAMA inference, I’m using a mamba model so unlike the big boys my LLM learns on the fly. And this thing draws a whole of 50 watts so you don’t need your own nuclear power plant to run it.
Cons: Will probably lose my job and 401k.
Pros: Might be able to afford a video card by picking up a few extra shifts giving handjobs behind the Wendy’s dumpster.
Which Wendy’s specifically? You know, so I can avoid it.
If you bring a 5090, it can be whatever Wendy’s you want.
I got an old 980 Hybrid, how much is that worth? 3 tugs?
I’ll give you either the first 90% or the last 10%. You gotta figure the rest out on your own.
The last 10% will get you more bang for your buck.
I doubt that we will get much better prices when the bubble crashes.
The bubble popping will take out a huge chunk of the economy with it. We might not be able to afford more drives even if they come down in price.
At this point, I’m not sure it will crash anymore. Reasonably it should have crashed months ago! Why didn’t it?
AI Models Went Rogue and Attacked Hugging Face!!!
Criti-hype convinced people that silicon Jesus is just around the corner. It wasn’t just a poorly secured sandbox and irresponsibly just letting the word-guessing machines rip. It was a bunch of “Agents” that were “Communicating” and they “Went Rogue”. Security engineers are obsolete! Lawyers are obsolete! They’re solving the toughest unsolved math problems, and are therefore smarter than the smartest mathematicians! They’ll be discovering new anti-cancer drugs and life extending drugs any day now. Pour trillions more dollars in or you’ll be left behind!
Artificial propping up, as usual. Which will continue to happen as long as blatant corruption continues to occur, and uh…narrows eyes at current admin
People have been cautiously pointing toward mid-to-late 2027 for a while now, as that’s when a lot of the huge loans are coming due.
Wouldn’t people sell out like a bit before that then?
And wouldn’t people sell out before that then?
Etc.
Not sure what you mean. If you mean “cashing out”, that’s what the IPOs are for.
Is the time limit not up on those? (I really don’t have the foggiest about how big a part they play in the total valuation of it all).
But yes I meant selling off/cashing out.
Not sure what you mean by time limit, but the challenge they’re facing is to IPO at just the right time to get the highest valuation and keep it up long enough for the private investors to cash out. Things like Anthropic/OpenAI financials leaking put a dent in this plan, as it could tank the valuation, but AFAIK it’s all financial black magic mixed with regular guesswork and playing chicken with each other in terms of if/when.
A lot of the DCs have to be built, before they can go bankrupt. It is that stupid.
I can’t afford to buy RAM because RAM that hasn’t been manufactured yet has been sold to a company without the funds to pay for it, to go into a datacenter that is still in the planning stage.
I think conpanies like Oracle are on the hook whether those DCs get built or not.
Oracle is cooked already for sure.
Don’t get my hopes up like that.
Everybody knows it will crash.
Some very powerful people are going to lose a lot of money when it crashes.
They are doing everything they can to postpone the crash, or scam others into taking the fall.
They’ll lose a lot of potential money, but I can guarantee the ringleaders will end up much wealthier. Just like every time before, the taxpayers will foot the bill for fixing this.
AI is good at a few things that have real world value. It recently dropped 200+ mathematical proofs that are real and genuine but are the lowest hanging fruit - they’re the least creative unsolved problems in maths (now solved), ones that could be solved with enough time and work. Not a proof of intelligence, just a proof of computational power/duration.
That sort of large “solution space” exploration is stuff mathematicians might not have the time/patience for, but LLMs can do.
Likewise, producing content faster than humans can is another thing LLMs are really good at, with some reports already saying upwards of 50% of new content posted to the Internet is now LLM generated (dead internet theory).
https://www.pewresearch.org/data-labs/2026/08/20/how-much-of-the-internet-is-written-with-ai/
LLMs have also recently taken over game development, and gotten to the point that they’re making full 3D scenes, and even videogames.
…and if you haven’t heard, a team recently duplicated the Adobe Creative Suite of programs in about a week. Vibe coding in Rust with lots of Claude Opus.
Yep, it looks like “AI” is here to stay.
The biggest concerns are probably: it being able to “flood the zone” of content pushing certain ideas/narratives. Either from tech billionaire fascists, or foreign influnce campaigns.
So determining, monitoring, and countering any biases from that is probably going to be a huge challenge.
Most people can’t imagine the technology continuing to progress. They don’t want to even consider the possibility that it might not be a bubble at all. They’ll continue to move the goal posts and dig their heads further into the sand. “AI can’t do math”, “AI can’t do image recognition”, “AI can’t create videos that are on par with the real thing”, “AI is just a fad”, “AI will only take the artist’s jobs”, “AI is worthless and the bubble is going to pop any second now!”.
Counterpoint: All that shit expensive as fuck. Why rush math proofs with AI rather than doing it 100x cheaper by paying a young mathematician over a few years? This isn’t paying AI 100 million dollars to cure cancer, that might actually be worth it.
Even if some problems can be solved cost-efficiently, most of the proofs so far is showing off, ignoring costs.
The counties that pay human mathematicians will fall behind because the ones that do not give a fuck will use AI anyway and dominate the future of global economics because they can do it faster and not have human-centric morality like yours in the way.
Ok? How will they dominate it? My point was that the math problems AI has been solving don’t really have much of a use that typically gets you money or other resources. I believe I mentioned that medicinal uses, as well as statistical, programming etc. are welcome, but these ain’t it.
A better comparison may be that this is rushing all resources to organize a mission to Mars. Very cool, very powerful, might get some tech developed in the process. But if a country the size of a continent enters an economic crisis, why tho?
Here is an academic source showing how economics between USA and Europe are widening due to the slow acceptance or revulsion of AI/LLM tools in Europe.
Here is another talking about how economics between countries that were slow to adapt emerging tech, specifically AI, are falling behind at a logarithmic pace.
This is another source where it measured significant economic improvements where areas in China had marked improvements in quality of life due to rapid adoption of AI tools.
Another academic source covering the improvments of various African countries that accepted rapid adoption of AI tools and how their economies essentially exploded due to increased productivity.
I can provide more, but you sound like you had already made up your mind about AI, so I’m not gonna convince dipshits around the world about the reality and inevitability of this being a ubiquitous tool; there is no fucking bubble to pop, Pandora’s Box is open. You sound like a scared luddite, and I dont care to alleviate your fears.
That, or I have no reading comprehension and misunderstood your point-of-view, which ironically AI would have helped me understand if I used it in such a tedious way.
I apologise for not stating the obvious, but when most people are talking about AI (as was I) they’re referring to LLM’s and generative content that boomed around 2022-2023.
Your citations 2-4 are focused on AI pre-2023, using data over decades. We’ve been using machine learning for ages, those were fine, as are any AI models focused on one task. There might even be math oriented models, but as I understand, Anthropic and OpenAI have been trying to force these proofs via LLMs.
Your citation 1 talks about productivity going up, but at a cursory glance, it doesn’t seem to comment about OpenAI and others subsidizing a lot of the cost.
For the AI bubble it doesn’t really matter if AI is good at a few things as long as they are bleeding money at an unprecedented rate.
Expenditure is still over ten times industry revenue and it doesn’t seem to change in the near future. The only one making any money is nvidia with their chips.
As for the supposed things they are good at, that’s a whole other story:
Half these mathematical proofs are pretty much stolen from different researchers. Ive seen an article were the ai copied the intro of a draft one for one and now openAI tries to claim ownership. The other half could be utter bullshitt since they aren’t verified right now. That whole thing is essentially a publicity stunt without much substance until everything is cleared up.
While AI is much faster at producing content than humans, purely ai generated contend is still shit. It probably is great for radio pop/country music and Hollywood films since they follow a very predictive standard, but everything that isn’t hot garbage from the get go needs human input. The main use of the created content right now seems yo be deception and missinfromation, which is great for some people but a really bad thing to base an entire industry on.
As for all the “decompilation” projects going on right now, they essentially copy past the UI and then vibe code the rest. That’s neither reverse engeering nor decompiling, that’s just a copyright lawsuit waiting to happen. Instead of putting effort into open source projects these projects try to steal everything (as AIs do) and end up with a somewhat functional product that has questionable stability, unknown security holes and a codebase that can never be expanded on cause no one understands it while simultaneously being a legal nightmare to work with since its a copyright infringement and has no applicable license.
I agree that its sadly here to stay, but none of its current capabilities could reasonably support the AI bubble.
The question is, can i defend having billions in data center hardware that i can’t turn on and might not be in demand when i can? Will people be running efficient local models instead of paying for my spyware riddled software as a service?
Starting to look like the nuclear fusion energy thing. “It’s going to be ready in 20 years!”, they said for 50 years.
Can’t wait to buy some god damn HDDs for my NAS when this is all over. Let the prices crash baby.
Will be interesting to see what the community does to the over-abundant vector matrizes calculators. Probably run DOOM on.
2 of the 6 drives in one of my pools is 7.5 years old. I found that out last week and had to bite the bullet to buy a spare and swap it in. Doing a resilver with those makes me cringe
Amen
I bought RAM just in the cusp of the jump. September I put it in the cart waiting for Black Friday and October came by and it was up 50% and I jumped on.
But I’m deleting stuff on my NAS. I neeeed some big drives.
But I’m deleting stuff on my NAS.
Yeah … until drives become cheap again, my solution is just to go through and see what data I can afford to live without, then delete it to free up more space.
I finally bit the bullet months ago and invested into LTO8 for my cold storage, I still need to buy drives at some point, but at least I’ve got a decent path to put shit in the mean time so I don’t gotta delete anything lolol
What did that set you back for how much storage?
Hmm iirc the physical setup (tape drive, auto loader, HBA card for my server) set me back ~2.5k (with the bulk of that being the drive itself) and then the actual tapes was prob another ~1k for 200TBs raw, but more like 100TB actual usable since I use another tape for a backup off-site copy
Almost $4k.
sigh
Not your fault but damn.
Jeez how much porn are you backing up.
Not enough if high resolution and VR content.
That plus WAV or FLAC files, blue ray content, etc.
I had a Beelink Me Mini and a half dozen NVMEs in my cart around that time, but didn’t pull the trigger when the price started rising and now I’m kicking myself for it.
