The lobster is out, read the room. You’re seeing articles of people stating local models are the future in tandem with hardware companies stating prices will go up for hardware and shortages will remain. This is a classic double tap from behind. It’s done. There were some good applications to it, which will continue. But overall it’s a toddler dropping their ice cream cone on the ground.
AI is the future. It is just that companies are rushing headlong into it, not properly planning nor preparing for it.
Sure, maybe one of them would be the next Google - but the vast majority are dinosaurs in a world to be filled with smoke and rats.
1 newspaper
Yeah, Bloomberg though. Not saying that they are great, but it’s among those that Wall Street bros read to decide what to buy and sell
“more clear that the real future of AI is small language and open-weight models running on local desktop computers” So where are we with those smaller models? I haven’t been following them. Are they any good?
Qwen3.8-flash-next is making waves since its near frontier or frontier (according to some i havent used it) that can be run on an ancient 8GB video card (12GB or more is recommended. 16GB is best.) with around 32GB of RAM using Strata.
It’s a mixture of experts MoE model, so while it’s around 66GB (depending on which quantization you decide to run) you’re only really using about 6B parameters per token processed, meaning the VRAM requirements are a lot lower.
I personally don’t see the value proposition in paying $4/1M tokens when i can generate around 30tok/s on 500W of power. I pay ¢13/kWh so to generate 1M output tokens takes me 9.25 hours and costs $0.60.
I’m also running an ancient video card and have a new one in the mail. If i got 200 tok/s even at higher power use i’m still a fraction of the cost of paying for API access and I want that GPU for other things regardless.
I feel lile medium and large businesses have a strong incentive to rent or buy inference using open weight models for a fraction the cost of paying OpenAI or Anthropic. Especially if you can pre load them with everything about your org to make them better suited for internal use.
The vast majority of uses can use a open weight locally run model. I think you would only want to pay for frontier models for really complex tasks with lots of steps, and even then there are usually ways to be smart about how you task agents with work to get better output and use less money.
The other issue is we are finding out that what you do in the AI cloud is logged. It can be subpoenaed, can be sent to police if it fails a safety check, or even stolen if the mathematicians fears are real. It is certainly being used to further train models…
Nice. I just wondered how they were coming along.
Right now, I have to manage which model does what. So I don’t. It’s the companies dime after all, and they clearly value speed over efficiency. But the model I directly talk to doesn’t need to be super high end. It could be well versed in which model is best for what task and farm put all of the work to the right model for the job. Seems like that alone could reduce cost a ton. But then start adding in cheaper local models and the price could fall a ton.
It would also be more accurate. Like if I had a small local model for interacting with a set of apis. It wouldn’t make the mistake of not considering that the api paginates results. The general model makes this mistake over and over.
So I agree this is the direction of the future. Just trying to figure out when we will get there.
investing in the wrong future
I’m assuming they mean “future” in the same way the stock market (ab)uses the word, because AI had its lower ribs removed on day one. There’s no future there.
I truly can’t see the bubble bursting at this point. They’ve won their race and built a technology so impressive it cannot be ignored or dropped. AI dominance is now a matter of nations (cyber) security. As soon as they’re any real chance of these labs failing they’ll be a tax payer bailout. I just hope it also results in nationalizing the leading AI labs so the models build on public knowledge can be forced into the public domain where they belong
They have not. Infact the opposite.
AI isn’t a race, It’s a marathon. You don’t sprint in marathons the way openAI and anthropic are. You get a 2 trillion dollar in debt just to run your business, then your business practice ain’t sustainable.
they’ll get a tax payer bailout
Under normal circumstances, yes. But geopolitically, it’s anything but normal. That’s the thing most people miss. USA interest rates are at an all time high. And Iran is doing it’s best to push it higher. Any large bailout would mean straining the us debt at a time it’s already in a bad position.
AI will live on but maybe as open weight models running on your laptop. There is no reason to believe that the cloud model is the only possible way for this to work. And there’s only so long that any company can lose billions of dollars a year.
I run a 122B model locally. I love local models. If you think they’re anywhere near as capable as the data center models you’ve either not tried them or are lying to yourself
I know that is the case today. But hardware and software will improve and the task sets they will be capable of will grow.
Are you using Qwen? Where do you find it useful/not useful?
Yeah Qwen 3.5 122B-A10B. It’s the most capable model I’ve found that I can run on my machine. It’s useful for simple tasks but honestly it’s around the border if correcting/checking its works is comparable effort to doing it myself. So I don’t really use it in reality it’s more a novelty. I’ve also messed around with Kimi K3 since the benchmarks said it was awesome. It’s definitely a league above Qwen but must be benchmaxxed to hell cuz it’s not even close to the same league as opus (I can’t speak to OpenAI models as my team has a Claude account so those are the paid models I know)
qwen flash is great. Privacy/sovereignty and protection from others stealing your data is very valuable and equivalence to frontier 3-4 months ago is very impressive. Your point is correct, and larger models have the capacity to solve problems with more general knowledge baked in, where smaller models simply can’t find critical information from search, or develop a strategy as well if they are clueless, and just repeatedly randomly guessing.
Totally agree on the privacy sovereignty front. I suspect in the long run personal tasks will be handled by local models for this reason. I guess I’m thinking more on the scale of institutional compute. Just like every large university has their own super compute cluster I expect them to want their own AI instance (I know companies like Amazon already have such a thing) and in those instances the gap in skill really matters and these institutions are willing to pay a fair amount for the best out there.
Also I can’t speak on equivalence with frontier models 3-4 months ago because I didn’t have access to paid models back then but if that’s true and not just bench maxxing then it means the moat is widening not shrinking. Which suggests these labs are in a good position to make themselves an asset the US cannot afford to lose despite their comically costly business model.
A bubble burst is not necessarily ‘the end’ of a tech. The internet bubble burst back in 2000, and here we are.
A lot of the financial burden is offloaded to SPVs that may well be allowed to fail, as the AI companies don’t directly carry the liabilities. They certainly carry enough of them, but they have done a lot to manage to leave other entitites holding the bag. Government may happily let Anthropic fold if it came to it, because Trump is all pissy with them. SpaceX would depend on whether Trump is pissed or loving Elon in the moment of decision. Hell if Trump decided that ‘TrumpAI’ was a possible concept he’d probably let them all burn for the chance of his grift picking up the pieces.
Can you explain more what you mean by this? For a bubble like the internet I can understand the giants failing and new companies picking up the pieces because the technology was already available to everyone so there was no real moat for new companies entering the space which feels quite different from the AI situation no? I’ll admit I wasn’t alive in 2000 so I don’t know exactly how all that went down.
There is this insane amount of investment in AI built around LLMs that threatens the entire world. People worry about the bubble bursting but think AI is the ant’s pants without considering that the LLM is a dead end. None of the AI companies are going to achieve Artificial Intelligence. That’s why they will actually implode, their systems aren’t intelligent.
well, they are being trained my the management, and as we all know these types wan’t 150% of the benefit with 25% of the work.
I mean, they are artificial intelligence by pretty much any normal definition of that term. So is Battle Chess from the 1980s. So is bank fraud detection. So is Conway’s Game of Life in its most basic definition.
Will they develop “super intelligence”, etc? No
I fear that yes they are not good enough to actually replace labor, but good enough to cause unprecedented chaos. Under capitalism, at least.
I think that’s the wrong conclusion. I agree they won’t achieve actual intelligence, but I see no reason why that would make them implode. Current tech is still going to be something companies pay a lot for.
It wont be a real AI, but itll be just enough to kill off a lot of the software industry.
I’m genuinely curious what “real AI” is to you?
A true Turing Machine, obviously
One that when given input will give the correct output, every time, without fault or stalling, in a timeframe that is quick
A Turing machine has nothing to do with AI. Pretty much any computer today can emulate one, the only thing lacking is infinite memory.
Real ai to me would mean true agency. And an entity which has its own goals and relationship to its world.
Yeah. I wonder if it might be difficult to identify something that reaches that point? I suppose genuine creativity might be an indication of intelligence? Or a desire to reprogram itself? I dunno, I’ve recently read Peter Watts blindsight and it kind of made me reconsider the relationship between consciousness and intelligence.
If someone truly believes LLMs are sentient, then they must under no circumstances prompt them even once; that would be murder.
Please explain without sounding crazy.
An LLM is a stateless stochastic function ran over your prompt (plus whatever system prompt) as input.
If you genuinely believe that the LLM is conscious, that means every time the function stops running (i.e. stops producing output tokens), the consciousness ends.
This is far closer to death than to, say, sleep, because prompting again doesn’t “wake” the LLM, it just runs the function again, from the top; as if a digital twin was given a transcript of the previous conversation and asked to take it from here.
(Obviously, “murder” was somewhat tongue-in-cheek, and this doesn’t even remotely meet any legal definition of murder. I was mostly poking fun at people who think LLMs are alive, without thinking through the implications. Personally, I think the idea that LLMs are alive/conscious/sentient is very often linked to not understanding that LLMs are not, in fact, a constantly-running ghost in the shell, thinking about and remembering things in the background.)
what software industry? React developer industry?
We’ve basically got to wait for Anthropic and OpenAI to actually IPO before any of this can get started properly. Once there’s a public expectation for profit from the two biggest AI labs, then the market will start acting against them. We can hope, at least.
Once there’s a public expectation for profit from the two biggest AI labs
I don’t think that it’s necessarily the order things will go.
Anthropic and OpenAI have bills to pay, including obligations to pay a ton of money to partners/vendors in the coming years. They need cash to do that. And the cash can come from 3 main places: revenue from customers, loans from lenders (usually in the form of issuing bonds), and issuing new stock/equity to raise funds.
The IPOs are basically their last chance to raise funds by issuing new equity. Even if each of them raise $100 billion, that’s probably not enough to hit self sustaining profitability. As lenders start to get more skeptical, in a macroeconomic environment with high interest rates and high energy prices, they start to run into a very real possibility that they hit illiquidity issues, even if their core business model actually works by then (and I don’t think it will). There’s just not enough runway for them, even in the most optimistic long term projections.
deleted by creator
Some datacenters could be repurposed as traditional datacenters focused on networking and cloud storage, infrastructure.
Look, none of use should be cheering for a full on pop of the buble. It will hurt all industries. And we know the oligarchy has insulated themselves in ways that will ensure they feel very little pain. It will be all the workers who will feel it. As for data centers, I don’t think we need the bubble to burst to fight that. Lots of jurisdictions are already banning them. That is what we should be cheering for.
While I am not cheering for the bubble to pop, the longer it goes on the worse it will be so the sooner it happens the better.
I dunno. The market affect is mostly perception. I mean why should unrelated stocks suffer when AI companies fall, yet they do. So I don’t think the length of time actually impacts the severity of the crash. But I’m no economist. That’s more of a “I don’t see why”, rather than a “no it won’t”
The main issue is that increasingly AI companies are making up a big portion of index funds which are required by law to be purchased by mutual fund managers and thus part of retirement accounts, so value will be evaporated out of retirees accounts. The larger the bubble the more the impact.
Also note that the only reason the US is not in recession right now is due to AI market “growth” so the GDP correction that follows will cause companies who have invested in the technology (not just AI companies, but most of the Fortune 500) to contract. The fallout will be unpredictable, but the most likely scenario is layoffs. Further, the Fed will be reluctant to cut interest rates, because inflation is extremely high due to tariffs, so the one lever the government has to stave off the recession has been compromised by the current administration. So the recession will be deeper and worse with a longer recovery than it would otherwise have.So why care? Because grandma’s retirement will disappear, a bunch of people will be laid off and the levers available to mitigate the impact have already been pulled. Oh and inflation will get worse when our government bails out AI tech bros to the tune of $1T.
So I agree on the impact, and on the why care. But I’m not seeing why waiting long makes it worse. Based on what you said, waiting longer could help if it can hold out until this tariff war is over (one way or the other). Then it might be possible to lower interest rates to soften the impact. If it can hold out for a whole new administration even, that could help a lot. Though that is also what would likely trigger it, as a new administration will likely restore regulations that have been relaxed, increasing costs to business. But I would rather pay that price at that time with that administration than this one.
The bubble is growing over time. It will have a more severe impact in two years than it would if it happened tomorrow.
There is no AI bubble, just like there is no housing bubble. It’s here to stay. This is the sad truth.
An economic bubble doesn’t require that the assets suddenly disappear upon the bubble popping.
The 2008 financial crisis was caused in part by the popping of a real estate bubble in the US. Those properties didn’t magically evaporate after the bubble popped, though many families couldn’t afford to keep their homes.
In the same manner, nobody is really suggesting that LLMs will suddenly disappear after the AI bubble pops, only that they will probably be proven to be significantly less profitable and scale back significantly.
In the same manner, nobody is really suggesting that LLMs will suddenly disappear after the AI bubble pops
There is a significant portion of AI haters who think exactly that, for some reason.
I mean, it’s likely that a good chunk of these hyperscale data centers and some of these big corporate LLMs will run out of cash and close shop. We’ll probably see a big pivot to much more efficient models that are local and/or purpose-built, not unlike what the Chinese are already doing.
Yeah dude, there was never a housing bubble and it never burst. Sure thing.
https://en.wikipedia.org/wiki/Subprime_mortgage_crisis
Some of us here are old enough to remember living through it.
How’s those houses priced for ya nowadays bud?
My friend, I have never been in the market to buy a house so I literally could not tell you.
My friend let me tell ya. There still unaffordable. But who knows, I heard on the internet that the bubble might pop someday.
they will literally restore slavery before they let a strategic corporation fail.
The funniest part is that the thing that could actually make AI useful at scale might be what wrecks the investment thesis: cheap, local, open weight models. Wall Street funded a trillion dollar server farm to discover that sometimes the answer is a smaller model running on your laptop. The technology might win while the investors lose. Nevertheless, companies such as Google, Amazon, and Microsoft – as well as Nvidia and AMD – are a very long way from simply going under.
Why do you think they are starving consumer market for computer parts?
I think we are talking about the same thing. AI companies spend billions to monopolize compute, causing consumer hardware prices to go up.
lol you’re kidding, right? Have you looked at component prices lately? Have you missed the part about how basically all consumer electronics that use LPDDR are getting price hikes, because the demand for high performance memory has skyrocketed by an order of magnitude in the last few years?
He didn’t mean to question the premise that the consumer market is being starved, he was highlighting the reality of that and connecting it as being motivated by a fear of local models.
Admittedly, the question was ambiguous and could be taken either way, but the comment didn’t mention that consumer market was starved, his question brought it into the discussion.
Note that I don’t think that it is fear of local models personally, I think it’s fear of each other. I think cloud infrastructure providers already proved that even when people can locally host stuff, they mostly don’t. So many consumer electronics could meet their promises all locally, but still force the users to connect to their internet services. The tech market has already proven time and time again they are suckers for subscription eshitification, no need to panic over it being different for AI.
Nevertheless, companies such as Google, Amazon, and Microsoft – as well as Nvidia and AMD – are a very long way from simply going under.
Not really, once you look at the startups like OpenAI they have enormous expenses that depend on constant debt and investment funding, and the cost and risk of sustaining that debt can easily reach a point where they struggle to get new investors to foot the bill.
The whole market is based on a belief in a huge pot of gold some time in the future, and it is driven by unfathomable amounts of debt to keep the machinery running towards that goal. Once the belief in the goal getting reached gets broken, the change in momentum breaks the whole system.
That includes the big players, huge amounts of their balance sheets, including claims of revenue, are circular trades between themselves. Even google claims revenue from asset price growth in their AI subsidiaries.
Once it shifts from investment to repaying bad debts the numbers will look completely different for all the big companies involved. We’re talking potentially bail outs basically across the board
The term bailout needs to stop being used and normalized. This is nothing like the housing market and audio industry crash.
These rich fucks don’t need socialism for the rich yet again, they need to eat their losses.
Google, Amazon and Microsoft will be fine.
Oracle, though. It’s possible they’ll actually die. And wouldn’t that be tragic.
I would comment about Oracle, but I don’t want legal bills.
Fuck 'em with a rusty rake
I doubt it. Government would step in and bail them out.
yep, with our money.
This isn’t going to happen. They’re not the banks, and they’re pretty universally hated in the country.
But their owners didn’t kiss the ring for nothing.
When the AI bubble pops it’s going to take down the banks.
There is a lot of private credit dollars loaned to data centers and the banks are heavily invested in private credit.
It’s a house of cards priced for perfection just waiting for one of the big 5 to have a slightly off quarter.
And whoever scavenges their corpse will milk their
customersprisoners even harder!they wont be fine, they will suffer heavy lossess but will survive, likely just by laying off hordes tech people, and even downsizing thier warehouses. i think nvidia will get off mostly less harmed though, since they sold thier products years ahead. hopefully it takes down skydance with it.
I think NVIDIA has a higher chance of survival because they’re the ones actually producing something tangible (GPUs) that is still needed even if openweight models prevail.
they’re doing far more than just that, the way jensen has been able to structure all almost all of these deals essentially offloading all the risk. they sell essentially lease the hardware, ontop of huge chunk of premium up front, ontop of paying for compute itself, ontop of offloading any actual buildout cost onto any greedy schmuck willimg to pay the tab. nvda will be completely fine (though stock price will obviously correct as bubble-levels income drops)
At the same time, they are doing extreme price gouging with little improvement between generations since h200. Huawei actually has better $/performance ratio and accelerating faster. Others have better inference solution, with AMD doing better on $/training performance too. The fact that Nvidia is the one propping up the AI bubble, sinking their cashflow into dead ends, makes it a bad stock too. Peak NVidia BS got exposed in an Australian IPO this week. It was too absurd to get anyone to buy and got cancelled. Heavily backed by NVidia.
shovels, gold, etc
Oracle, though. It’s possible they’ll actually die. And wouldn’t that be tragic.
Oh the humanity
Too big to fail? I think I heard that once.
I just can’t see things continuing like this, there has to be a plateau reached at some point. Higher or lower. Exponential growth isn’t sustainable in anything.
As a financial analyst, Klement’s take on the AI economy has been notably grim — back in May, he penned an article in Reuters arguing that institutional investors shouldn’t be asking “whether the AI boom will end, but what will happen to markets when it does.” Now, his live interview on Bloomberg‘s flagship news channel underscores the degree to which the AI bubble is becoming impossible for mainstream commentators to ignore […]
What’s interesting to me is that people who run Bloomberg have a huge advantage in manipulating the market.
In fact, the story behind the AI boom really has to be how the business press operates, always enabling the upper class to be richer.
Of course, if the Iraq war is any example, there will be a few confessions and nothing will change.
Yes any time now. Just like in 2022, 2023, 2024, 2025…
Nvidia is not even overpriced. If anything, it’s underpriced because it’s a 5 trillion dollar company and the market doesn’t believe it will be able to keep growing.
I dont believe this technology will become anything actually smart. It’s still just text prediction. But it’s good enough for a lot of usecases.
Is that why someone on Thursday just spent $21 million on 100,000 puts at $180 of NVDA?
That’s cool. Link to it please. Some prediction market?
It isn’t profitable at current pricing, and it will have lower adoption at higher pricing.
The technology isn’t going anywhere, when the bubble pops and hardware prices come down, local open weight models will be sufficient and the hardware to run them will be cheaper than cloud models.
In the short term, however, it’ll be a shit show.
I really wish you guys who believe this would be able to go forward in time and see what happened. :)
Do you have a reason to believe what I said is unbelievable?
It’s their own numbers…
Maybe, but im convinced it will work just like any other technology that has arrived before it.
Currently we are in the hype stage where so called experts are warning the technology can kill humanity and other silly things.
But this will pass. The technology will continue to be very useful to both companies and individuals and because of this, pricing will adapt. This means less price for the consumers as number of users and competition goes up.
We will have a society where Ai is used to take our orders, find our information, do a lot of office job tasks. It will become just as common everywhere as screens and phones.
And people say it’s not going to be profitable? That’s extreamly improbable. Every technology that is used by everyone is highly profitable, it’s just a matter of which companies will win the market and which won’t.
There won’t be any bubble popping because it’s not a bubble. It’s a new technology, like phones and laptops once were too. I had one of the first smart phones actually and I watched that market explode over the coming decade, with every human buying a smart phone. That market is now stagnated, more of less. But Ai has only just begun.
The problem is not that it’s a useful tool for some applications. The problem is the insane amount of money thrown at this which is not possible to repay unless their product achieves runaway AGI and can actually cause these doomsday scenarios. They have no path to profitability. Like, none.
These companies are so rich and already have existing revenue in billions, so even if the Ai bet doesn’t work out, they can simply pay off their debt using their ordinary revenue.
It would disappoint the stock market, sure, but I doubt that it would create anything more problematic than maybe a 20% drop in the stock, temporarily. And that would be because investors would know that the future will not contain the promised profits. But the companies would still be fine, still be able to pay their debts, and still be able to run their business as usual.
Its not like in dot com where companies without revenue got to borrow tons of money they could never pay back.
Its not like in dot com where companies without revenue got to borrow tons of money they could never pay back.
It’s exactly that, but much worse. These companies do not make money. They lose tons of money every quarter. They have revenue but it doesn’t even cover their operations cost. But that’s not even the expensive part; training new models is hideously pricey and that will not change. And they must keep training new models lest they “fall behind” in this race to the bottom. OpenAI owes a Trillion and a half dollars to their investors. They have at least a trillion in obligations to compute providers. They keep taking more investor money, most recently with SoftBank who sold junk bonds to raise it.
Not to worry, OpenAI may lose money on every sale, but they make up for it in volume!
As how my local AI would put it, “I would like to gently push back”:
There is an AI bubble. Just as the Dotcom Bubble, it won’t kill the technology nor slow advancement - it simply selects against companies that are not fiscally responsible or intelligent about their agenda.
Personally, in this paragraph:
The technology isn’t going anywhere, when the bubble pops and hardware prices come down, local open weight models will be sufficient and the hardware to run them will be cheaper than cloud models.
I honestly believe that the AI hype is just the justification for the price hikes on hardware, the point is removing access to capable hardware for all of us so that they can have most, if not all, control over the population.
I certainly hope you’re right, but all signs say otherwise, in my opinion anyway.
Get off YouTube! Your thinking has been poisened with conspiracy mindset.
That may very well be the case. Now, is it really a conspiracy when everything points to it, or is it a justified suspicion?
Conspiracies exist. Many are well documented, or at least documented enough to call out specific people for specific behavior.
A conspiracy mindset ignores those, and instead suspects conspiracy everywhere else, and usually without substantial evidence. Easy instances are detectable when the sentence goes they do something to harm us in a vaguely related way.
The latter is all over YouTube, and what influencers make money with. The former is sometimes reposted on YouTube after some journalists publish it in a serious media outlet.
Well… We all go forward in time. That’s kinda how time works (generally).
Unfortunately we can’t go there and back. :)
I’m so burnt out and jaded on AI that I don’t think I’ll ever use it for any reason ever. I just wish we could put the cat back in the bag. I know that will never happen.
AI could be used to save our planet from catastrophic climate change, to problem solve ways to improve things, to do hard labor so that people don’t have to… but nah let’s use it to make assholes with too much money get more money! That’s fantastic
its doing exactly opposite of that, i dont think it will save anything, its already catastrophic enough to have all these polluting datacenters in every country.
They didn’t listen to the scientists.
Somehow I think AI would go from “Super Intelligence” to “that libtard tree hugging computer made of parts from commie Chy-na” as soon as it told them to do something legit that would inconvenience them or disagree with their narrative.
We already got solutions. What’s missing is the courage and incentives to actually adopt them by political leaders and the larger public. If AI would replace our politicians maybe it could help, otherwise it’s just the latest proof that there’s plenty of money around to transform the economy into whatever we want it to be. I guess we prefer chat bots over carbon neutral.
