The current bubble is getting scary. When this thing pops the blast it’s going to be a real mess. The big tech firms will hurt, fire some execs in a show of “making changes,” do a bunch of layoffs across “AI” teams to show the market they’re pivoting and getting costs in order, and move on. The startups ecosystem will suffer extensive and catastrophic damage.
The funding ecosystem will be set back years as this wipes out a bunch of VCs and investors. Some of the “AI startups” will be sold in fire sales for parts so investors can at least minimize losses and most of the rest will vaporize, the likes of which we haven’t seen since financial firms imploded in 2008.
I hope I’m wrong, but the most experienced trusted folks I know are already repositioning themselves to weather the upcoming storm.
I kind of hope you're right. Any "hyped" industry/sector is bound to eventually needing to get back to reality, and focus on things that actually work, rather than spraying and praying prototypes and over-hyping them.
The individuals and companies building real products that actually improve something will stick around, either as they are, or at least as ideas, and most of the interesting stuff tends to happen when the hype dies down, as the builders continue as they were, but all the rest of the riffraff disappears.
There will still be a community and the ideas won't magically disappear, just smaller and more focused, which to me sounds like a much needed improvement over the current state of things.
I couldn't care less if big tech gets knocked down a peg, but in many quarters the AI boom is what's keeping the lights on. A market correction of that magnitude would mean a lot of pain for a lot of normal people, it's not exactly something I'm cheering on...
Yeah, it's like an infection. Do you want to deal with it now, when it's painful and you'll lose your finger or when you risk your arm, or risk dying of Sepsis. You can put off the pain but that will make the outcome worse.
As someone who've jumped on previous hype-trains, most of us who get involved with "frontiers" know it's over-hyped, have more realistic perspective of it and are more-or-less ready for what eventually will come.
I'm sure most people heavily invested into AI (energy, money or time-wise) know the consequences of the bet they've done, they're not exactly just trying to earn a living.
If the crash/bubble-pop would have change of impacting actual working class folks outside of the AI bubble, I'd agree with you. But I don't think the bubble is so large, the people impacted will be people who willfully made a risky bet.
I don’t know. I’ve read the AI bubble/spending is so big (new data centers, power plants, etc) that it’s propping up GDP. If that all went away plenty of normal people could get hurt. Still, probably better sooner than later. Well, maybe after the power plants are done so we have excess capacity.
But not knocking it down will also mean a lot of pain for the people? Just look at the unsustainable electricity demand and skyrocketing prices big tech is pushing upon the small folk due to their explosive needs for more and more big data centers.
I think it would be better to have a bust now when the actual sums that have already been spend are relatively modest (~100bn?) rather than the $1tn over the next couple of years people are talking about.
That return to reality has happened many times in the past. It has happened several times before with AI! In every case with the worthless stuff the bubble was propping up we also lost a lot of useful things too and it generally took a decade for the people who had those useful things that the bubble could help with to realize they shouldn't have stopped investing with everyone else when the fad changed.
Well we’re gonna get, and need, a big correction. I just hope it’s a “big correction” and not a financial implosion that sets the startup ecosystem back a generation.
Hard disagree. The correction you envisage to something smaller and more focused won't happen because this is also a race between nation states. The bubble is in the startups, and you're right to suggest many are doomed. But the AI giants have strategic national importance now, and won't be denied funding and resources while their foreign rivals continue to grow.
In my naïve of opinion, AI is not necessarily about coding or making things better, it's more about creating a system to control the message to the masses. Just like Facebook controlled the message during the Clinton/Trump campaigns, AI is going to control the message for the vast majority of easily swayed people. Especially now since it is very hard to detect between real and AI generated persons/data (see Tilly Norwood).
This relates directly from your post about nation states. Those who control the message control the people.
The only upside here is if a few notables end up penny-less and dressed in orange. The mockery of solvent business practices, greed and theft is insane in this bubble. From copyrighted material to accounting fraud, I'm sure the fallout motion pictures will be entertaining some day.
What about the rest of the economy? Deutsche Bank recently said that the AI hype is the only thing holding the US stock market together. And if that crashes, it tends to ripple world wide.
Scary stuff, at least to someone who doesn't know all that much about market resilience. With what little I know, I'm hoping for a soft pop with slow deflation. But big tech seems to just pump harder right now.
Any big bust will have broader market implications, but I don’t think this has the broad systemic impact that the financial crisis did. It will likely be really ugly for those caught up in it, but a news story and minor blip to the 401k of everyone else. If you are well diversified keep your head down and keep going. If you’re a VC that heavily invested in AI, I suggest preparing for a Cat 5 hurricane now.
The biggest impact the average person will see is that the days of VC funded cheap AI will be over. If they want to use AI to cheat on their essay they’ll need to pay a lot more.
I would caution that people who think they've diversified are not. A huge portion of the S&P 500 is the top handful of companies, so an index fund may mostly end up heavily invested in those firms.
Some people have shifted investments around to fix that, but I'm sure many people will be quite surprised.
Agreed. If you look at what's in the S&P 500, 7.49% is nVidia, 6.33% is Microsoft, 4.88% is Google, 3.86% is Amazon, and 2.95% is Meta. That's over 25% of the S&P 500 in companies that will see their share prices fall heavily if there's yet another AI winter.
Under diversification, as you are describing, is the first ingredient…
I suspect the second ingredient is US boomers right on the edge of retirement or otherwise shifting to fixed income… And a normal correction causes all of them to divest simultaneously from equities.
There is a potential for a stampede out of US equities.
NVDIA has plenty of other business and will not drop to 0. Even if they drop 75%, than is still only a 2-3% (the math on percentages is weird and I don't feel like doing it - and I shouldn't because a 75% drop is just a random number and so I have no significant figures to work with) drop in my 401k. Of course other companies will be hit as well, but still my 401k will not drop and more than any other stock market crash, and there is every reason to think it will recover. It might stop my plans to retire in the next 10 years (but they were of questionable realism anyway), but any longer than that and odds are it will recover.
I'd say you are underestimating the industries being propped up by the AI Boom.
To name a few examples - Construction (the industry pivoted to DCs) [0], the entire hardware industry (from personal experience CHIPS act recipients pivoted to AI-hype because of slow reimbursement rates), whatever tech hiring that even exists in the US at this point is itself looped into and justified by the AI boom or capital that is 1-2 degrees removed from the AI boom, the entire energy industry is tied to the AI boom [1], and the renewables industry in the US is increasingly AI-washed now as well [2] because of the ending of IRA and IIJA disbursements under Trump.
This wouldn't be a 2008 sized event, but it would be a bloodbath comparable to the Telecom Bust (2001) and the Dot Com Bust (2001). I would trust that every HNer would be worried about such a situation because of how existential a risk it would be for the entire tech industry - which employs most people on this forum.
You're hoping for a full on market crash? I don't mind AI hype companies crashing, would probably low key enjoy it with popcorn. But from what I understand, it's not that localised.
Yeah it will be funny. I couldn't care less. My tea and tobacco will still be just as cheap as it ever was. I'll still be playing my lute on the porch, unaffected.
My startup is already dead in the water because I built it in an ADHD-fritz and only did it to spite a YC company, and in that regard I succeeded wildly because they threatened to sue me. Haha! I'm working on much cooler shit right now. Sick own though, checking my profile and all that.
It's not supposed to be a burn. My point is being flippant about a potential market crash under the assumption that you can somehow weather it out is extremely unrealistic unless you are a HNWI.
It’s pretty scary. According to Barron’s, MicroStrategy, a bitcoin treasury company, alone makes up about 5% of the U.S. convertible bond market. That’s remarkable given that it isn’t a typical tech or biotech growth company issuing convertibles, but essentially a Bitcoin treasury company.
> The funding ecosystem will be set back years as this wipes out a bunch of VCs and investors
I don’t think this is likely. We’ve seen VCs overextend into social networks, “sharing economy”, B2BSAAS, machine learning, developer tools, video games and crypto, in the last 15 years, and they have very little to show for it. Something new will come along and they’ll invest in that.
VCs typically invest money invested in them, so if that dries up, regardless of what dream they want to chase, they can't. The hypes you list all happened, but I don't think I've seen anything in the last 20 years that even comes close to the current AI hype.
I think the debt is what makes the risk systemic. Sure it can be overvalued, but alone that won't cause a generationally painful economic meltdown. It's the spending and debt that make something that is a bubble into something truly dangerous. All the exotic private credit and structured finance that is powering this thing along with the lack of any viable revenue stream to keep up with the debt plus interest that make this thing so dangerous. Sure maybe Meta can take a huge hit and limp away (it's still gonna be very painful for them and people who own a lot of its stock) but can OpenAI, can CoreWeave or any of the firms that lent to them? It's a domino effect. The fact that Meta is doing this is a huge red flag. The problem is the market is rewarding this endless cash burn without any way to generate the appropriate revenue. Once reality catches up there are a ton of knock on effects.
The same way they always do every single year: they re balance their portfolio so they are never heavily invested in any one thing.
If you believe in AI and want to bet strongly in it - which some experienced folks do - you take 5% of your portfolio and bet that in AI. The other 95% is invested in a diversified portfolio.
There are many inexperienced investors. Anyone can ride a bubble up and make a lot of money. There is no reason to think you can call the top of a bubble (or if there is a bubble!) consistently enough to bet on it.
I'm not 100% sure how YC works, but generally places like that are invested in by people who use it as the risk portion of the portfolio. So you while YC might be 100% in AI, the people behind YC or only 5% in and so it works out. If you have 100% in any company then you are diversified and need to fix your portfolio.
It is generally best for a company to specialize in something they do well. It is possible that YC has picked winners because their specialize in this, and thus even though the AI bubble collapses they are okay. Who knows - but this is something specialists can do in some cases
> The startups ecosystem will suffer extensive and catastrophic damage.
These a feature, not a bug of how the startup ecosystem works.
> The funding ecosystem will be set back years as this wipes out a bunch of VCs and investors.
Once again, a feature. There's too many VCs and too many funds.
> the likes of which we haven’t seen since financial firms imploded in 2008.
Institutional financials firms dwarf VC private capital. This alarmist comment makes it sound like a nuke going off in a Nevada desert is going to kill a major US city.
META free cashflow last year : $20bn. Cash on Hand: $47bn. "Worrying" Debt: $15bn sought.
ORCL, the other company they're talking about: $20bn in Cash from Operations, $21bn in capital expenditures, ORCL Cash on Hand: $11bn. ORCL's recent debt flotation: $18bn.
ORCL has 40 years to pay back; demand was reportedly $88bn for the offering. I imagine pricing was close to T-bills.
These flotations posit that demand for compute will continue to increase over the next 40 years, and that infra providers who can get there early will do better than Treasuries.
Calling it a bubble just because the numbers are big is the weakest of financial journalism. Now, do you think inference and datacenter demand will drop? If so, it's worth asking if and when these datacenter will pay, and if they don't, who will take the hit. That would be useful analysis.
I'm pro these plays -- right now inference has an 80% margin; that's after paying the fully capitalized costs of datacenters + compute + the datacenter margin. To the extent a company controls its own inference stack and can do so with a 5% cost of capital, they should do it.
> Calling it a bubble just because the numbers are big is the weakest of financial journalism
Yeah, exactly. The current AI investment wave could be a bubble. Or not. If anyone knows, there are millions to be made in the stock market. The answer depends on the actual expectations for those investments and how actual business metrics are tracking them.
But pointing to ambiguous “evidence” just adds noise. If someone screams fire in a movie theater journalist should report if there’s a fire or at least smoke, not write articles showing concern about how flammable all the furniture is
The thing is, the models do work. They add value to me each and every day, to a degree almost no other tech has done before.
But that doesn't take away the fact that this is extremely expensive stuff (not for me, but for the companies pushing the envelope), far too expensive. And it is really taking its toll on other resources, like electricity.
I think the thing that most of the folks miss is exactly that. AI is adding value and it’s here to stay. Absolutely the OpenAI story is looking concerning but I am not sure it is this doomsday AI winter.
Both can be true at the same time. Similar to the early days of the Internet, the dot-com bubble eventually popped, but the Internet (and dot-coms, for that matter) didn't go away.
What people are saying is that this mad race to throw cash at anything that has "AI" in it will eventually stop, and what will remain are the useful things.
Yeah, I saw "AI winter" mentioned elsewhere in the thread...
IMO there is a real qualitative difference between AI and crypto in terms of the durable impact it's going to have on the world. Does that mean I've bought into the AI hype? Maybe. But I think the signs are there.
AI winters are a recurring phenomenon, not a myth, and, like the dotcom bust, involve a collapsing hype bubble, reductions in focussed speculative investment in the field, but the technologies that were big during the preceding hype cycle continuing to be important, and develop, though in the case of AI winters often they stop being thought of as AI and just get referred to with a name for the specific technology (often a different one than the main one they were known by in the hype cycle, e.g. “expert systems” from th blast hype cycle are largely “business rules engines” now.)
Googling, there seem to have been two AI winters, the first (late 1970s - early 1980s) when people first figured AI was overhyped, and the second (late 1980s - early 1990s) with the collapse of expert systems. I don't think we are about to get one now - more like AI spring leading to AGI summer.
If revenue doesn't catch up with the cost of developing and running these huge LLMs then the only way to avoid an AI winter is to find a way to make them way cheaper to develop.
I keep seeing everyone guessing what the margin is on inference. You say that it's largely break even. We have this person in the thread claim 80% margin (https://news.ycombinator.com/item?id=45462442).
Why would they be make believe? I am generalizing across multiple companies and simplifying it as much as possible as we won’t know all the cost buckets but we do know at current costs the bare inference cost of keeping the machines running is being covered.
But are you paying for them commensurate with the value they create? That's the problem. Tons of money being invested/borrowed on technology with not nearly enough revenue to justify it. Same thing that happened in the late 1990s.
Actually the problem is the same as all previous technology: is the company getting 100% benefit while employees get less than zero with lost jobs, lost negotiating power, lost unique skill marketability, etc
No question that neural networks can add value, as has been true for a long time. The question is, how much? The current expenditures indicate that the answer is, "trillions of dollars worth".
If you call it "neural networks" or "large language models" or "machine learning", this would sound absurd. Only by calling it "AI" (with the implication that it will achieve general intelligence) does this sound, for most people, possibly correct.
It's just gotten so out of hand that there are multiple levels of pain. There's all the debt financing / private credit deals on infrastructure costs that will be defaulted on. There are the losses that most people will feel when the overvalued shares drop. There is the pain of tech companies having spent so many resources with very little in terms of viable revenue streams. It's gonna hit a lot of places in the economy.
> "SPVs mean companies like Meta do not need to show the debt as their debt," Perkins writes in a note. He likens today's financing tactics to the subprime era when firms shifted risk off the books to reassure investors.
But it's not? If the SPV is doing its job as a limited liability entity, Meta should be bankruptcy remote from the SPV. Presumably accounting rules would look through that SPV if Meta was actually reachable. It doesn't make a good comparison to GFC where everyone was nominally on the hook but had that risk "insured" with CDSs, etc.
Perkins’ point is valid. It’s still shareholder assets and resources at the end of the day even if via a SPV. The banks did this sort of thing in the financial crisis with all sorts of subsidiaries and circular funding.
“Hey it’s OK that mortgage side business is going belly up but we bought that special insurance.”
“Cool, so who sells that insurance?”
“Um, actually I think we do via our other SPV”
“$&@!”
Not 100% analogy here but similar shenanigans where company A is taking out debt, to build a datacenter, to buy GPUs, to sell services to startup B that Company A gave cash to to buy said services. Startup B has no viable business model and implodes. Company A goes from looking like a genius to a financial nuclear waste dump almost overnight. It’s financial shell games all over again.
Isn't the problem the creditors exposed to the SPVs though?
Meta, and others might be shielded from it but there's someone on the other end of these debts, if the SPVs are accruing massive amounts of debt and go bankrupt there are lots of obligations that will go unfulfilled, bringing the whole house down.
It's quacking very similar to other accounting tricks, caham, financial engineering, to hide bad numbers somewhere else. CDSs and CDOs were a different mechanism with the same purpose: hide risk away into overly complex instruments.
> "SPVs mean companies like Meta do not need to show the debt as their debt," Perkins writes in a note. He likens today's financing tactics to the subprime era when firms shifted risk off the books to reassure investors.
The article doesn't explain how Meta are actually using SPVs. The suggestion is that the debt is in fact Meta's debt but the use of an SPV somehow means they don't need to show it on their balance sheet. Traditionally if you're using an SPV it genuinely means it's not your debt, in that you won't go bankrupt if the deal flops and the assets are underwater. (And they're not your assets, either.)
There are valid concerns about debt fuelling outsized investments in AI, but this article isn't really saying anything of substance. Plenty of non-bubbles attract significant debt investment as well.
Ultimately, debt is a way to transfer risk. It becomes an issue if it results in a lot of risk being transferred to parts of the economy that can't really afford it (like retail investors, or systemically important banks). As long as the people taking on the risk know what they're getting into and we won't need to bail them out if they lose their shirt, it's not really a problem.
That itself is not proof that AI is not here to stay. It’s a short minded world view. Value is absolutely being generated but of course not everywhere and I am certainly not defending valuations.
Value is being generated for owners. Not for me. I don't care at all about "value". I don't know what perspective you are thinking from. Are you rich? AI and crypto and every flavor of the year tech garbage are just siphons. Judging the tech on its own merits is missing the forest for the trees.
Dismissing an entire technology because you personally haven’t figured out how to make it useful is a cope, not an argument.
Am I rich? What are you even on about.
I can now spend a few dollars and classify a large dataset that would have taken me real time before. I can again spend a few dollars and turn unstructured data into structure. There absolutely is value to be found in the current gen of tech. Comparing it to crypto is a weak argument. Just because you don’t find value does not mean other organizations and people are not.
The only people I find harping on about some amorphous "value" floating around in the air are typically people with portfolios.
>I can again spend a few dollars and turn unstructured data into structure.
Saving 5 seconds on a 10 second task using an infinite amount more energy is not "value", it's garbage propped up by VC money and regular people are going to be paying for it when it inevitably goes tits up. Like we've paid for everything. There's your "value".
You’re confusing your frustration with some half-baked React scaffolding with the entire field. “Saving 5 seconds on a 10 second task” is just a strawman, there are workloads where it saves weeks, even months. If you can’t see that, fine, but don’t pretend your blind spot is universal truth.
And spare me the VC boogeyman. The fact that capital flows into bad bets doesn’t erase the actual wins. We’ve had garbage propped up before, railroads, dot-coms, mobile apps but guess what? Railroads exist, the internet exists, mobile exists. Same pattern here.
I cannot say for your work but for classification steps and data structuring it’s quite accurate and this is with regular testing. I cannot speak for your work but for mine and folks in adjacent industries, LLM are fantastic and adding a lot of value to our workflows. You’re honestly holding on to this dead idea that LLM outputs are full of hallucinations. Throwaway account with throwaway comment.
It's "quite accurate" which is not acceptable for almost all relevant tasks is a business context. Somebody needs to manually check everything. Almost no time is saved.
Talking as someone who has built many small OpenAI integrations aka wrappers in business apps.
I would pay good money to be this green again. I talk about regular people, and you revert back to spreadsheet efficiency. Classic tech guy response, just cannot see a bigger picture.
“I’d pay good money to be this green again” just reads as bitterness dressed up as wisdom. You’ve offered zero substance beyond sneering at anyone who finds use in the tools. If your whole argument is “I don’t like it so it’s worthless,” that’s not insight, it’s just cope.
- ChatGPT still lags for me daily and I'm on a PRO subscription.
- SORA is usable but extremely slow and constrained.
- Image gen same thing, but getting a bit quicker and less constrained.
This is with agents not having taken off yet, and the vast portion of the world economy not interacting with LLM Ai? Agents alone will require ungodly amounts of compute.
What exactly are you referring to? Their UI seems to have troubles to load the list of conversations since like a week back, otherwise it seems fine? I mean, UX could be a lot better, but I'm experiencing anything I'd describe as "lag".
Yeah, especially considering it's only fetching ~10 items at a time :) I'm guessing some index somewhere is wrong or something, as I probably have a couple of conversations through the years of using it almost daily...
Dang. My kids are just getting the hang of using ChatGPT as a way to generate random quizzes and additional homework problems; to the point where both of them just automatically use the AI to "fill in the gaps" each evening. (Prior to AI, additional HW was met with less enthusiasm than you'd think!) I told my wife that the billionaire class spent the last few decades cheating the tax system (and, thus, the education system) just to weirdly dump a trillion dollars into the AI ecosystem to build the world's finest tutors. I really hope that doesn't Go Away™.
Even if OpenAI folds, the open source stuff is good enough that someone will build a compelling tutor platform in their place. It probably already exists.
Don't worry about this. There's no way these tools are going away. If this bubble bursts it may wipe out the incentive to continue this frenzied race to build novel AI, but ChatGPT et al won't be shut down, and even if they were, the open-source LLMs comparable to the cutting-edge of 6 months ago will still be online and available. Plus, even if AI progress froze solid tomorrow, I think it would take decades before we'd start to anywhere-near-saturate the potential space of applications & use cases to really do the current tech level justice. (Also, even post-bubble, AI progress would not freeze solid, to put it mildly)
>Meta seeks $29 billion via private capital for its AI data center buildout.
This would be worrisome if they didn't have the cash to pay for it, but that's peanuts to Meta. It seems more like tax avoidance schemes than anything else.
The bubble will burst with a vengeance when big corporations and startups start deploying and utilizing LLM models on-premises instead of the "pay-as-you-go" trap.
B... bb... bbbut i thought we are so close to AGI and all the AI companies can easily make bazillions off of targeted advertising and and AI can do all my employees jobs for 1/1000th their salary and and...
The current bubble is getting scary. When this thing pops the blast it’s going to be a real mess. The big tech firms will hurt, fire some execs in a show of “making changes,” do a bunch of layoffs across “AI” teams to show the market they’re pivoting and getting costs in order, and move on. The startups ecosystem will suffer extensive and catastrophic damage.
The funding ecosystem will be set back years as this wipes out a bunch of VCs and investors. Some of the “AI startups” will be sold in fire sales for parts so investors can at least minimize losses and most of the rest will vaporize, the likes of which we haven’t seen since financial firms imploded in 2008.
I hope I’m wrong, but the most experienced trusted folks I know are already repositioning themselves to weather the upcoming storm.
> I hope I’m wrong
I kind of hope you're right. Any "hyped" industry/sector is bound to eventually needing to get back to reality, and focus on things that actually work, rather than spraying and praying prototypes and over-hyping them.
The individuals and companies building real products that actually improve something will stick around, either as they are, or at least as ideas, and most of the interesting stuff tends to happen when the hype dies down, as the builders continue as they were, but all the rest of the riffraff disappears.
There will still be a community and the ideas won't magically disappear, just smaller and more focused, which to me sounds like a much needed improvement over the current state of things.
> I kind of hope you're right.
I couldn't care less if big tech gets knocked down a peg, but in many quarters the AI boom is what's keeping the lights on. A market correction of that magnitude would mean a lot of pain for a lot of normal people, it's not exactly something I'm cheering on...
> A market correction of that magnitude would mean a lot of pain for a lot of normal people
A 1% increase in US unemployment results in thousands of excess deaths per year.
It is hard to be exact about the numbers, but the trend is clear and strong, see e.g. https://pmc.ncbi.nlm.nih.gov/articles/PMC3070776/
Continuing the masquerade will only make things worse
Yeah, it's like an infection. Do you want to deal with it now, when it's painful and you'll lose your finger or when you risk your arm, or risk dying of Sepsis. You can put off the pain but that will make the outcome worse.
As someone who've jumped on previous hype-trains, most of us who get involved with "frontiers" know it's over-hyped, have more realistic perspective of it and are more-or-less ready for what eventually will come.
I'm sure most people heavily invested into AI (energy, money or time-wise) know the consequences of the bet they've done, they're not exactly just trying to earn a living.
If the crash/bubble-pop would have change of impacting actual working class folks outside of the AI bubble, I'd agree with you. But I don't think the bubble is so large, the people impacted will be people who willfully made a risky bet.
I don’t know. I’ve read the AI bubble/spending is so big (new data centers, power plants, etc) that it’s propping up GDP. If that all went away plenty of normal people could get hurt. Still, probably better sooner than later. Well, maybe after the power plants are done so we have excess capacity.
But not knocking it down will also mean a lot of pain for the people? Just look at the unsustainable electricity demand and skyrocketing prices big tech is pushing upon the small folk due to their explosive needs for more and more big data centers.
If the hype doesn't line up with the fundamentals, then the bubble will have to burst sooner or later.
I think it would be better to have a bust now when the actual sums that have already been spend are relatively modest (~100bn?) rather than the $1tn over the next couple of years people are talking about.
Your comment made me think of this post:
https://www.palladiummag.com/2025/10/03/how-gdp-hides-indust...
That return to reality has happened many times in the past. It has happened several times before with AI! In every case with the worthless stuff the bubble was propping up we also lost a lot of useful things too and it generally took a decade for the people who had those useful things that the bubble could help with to realize they shouldn't have stopped investing with everyone else when the fad changed.
Well we’re gonna get, and need, a big correction. I just hope it’s a “big correction” and not a financial implosion that sets the startup ecosystem back a generation.
Hard disagree. The correction you envisage to something smaller and more focused won't happen because this is also a race between nation states. The bubble is in the startups, and you're right to suggest many are doomed. But the AI giants have strategic national importance now, and won't be denied funding and resources while their foreign rivals continue to grow.
In my naïve of opinion, AI is not necessarily about coding or making things better, it's more about creating a system to control the message to the masses. Just like Facebook controlled the message during the Clinton/Trump campaigns, AI is going to control the message for the vast majority of easily swayed people. Especially now since it is very hard to detect between real and AI generated persons/data (see Tilly Norwood).
This relates directly from your post about nation states. Those who control the message control the people.
The only upside here is if a few notables end up penny-less and dressed in orange. The mockery of solvent business practices, greed and theft is insane in this bubble. From copyrighted material to accounting fraud, I'm sure the fallout motion pictures will be entertaining some day.
What about the rest of the economy? Deutsche Bank recently said that the AI hype is the only thing holding the US stock market together. And if that crashes, it tends to ripple world wide.
Scary stuff, at least to someone who doesn't know all that much about market resilience. With what little I know, I'm hoping for a soft pop with slow deflation. But big tech seems to just pump harder right now.
Any big bust will have broader market implications, but I don’t think this has the broad systemic impact that the financial crisis did. It will likely be really ugly for those caught up in it, but a news story and minor blip to the 401k of everyone else. If you are well diversified keep your head down and keep going. If you’re a VC that heavily invested in AI, I suggest preparing for a Cat 5 hurricane now.
The biggest impact the average person will see is that the days of VC funded cheap AI will be over. If they want to use AI to cheat on their essay they’ll need to pay a lot more.
I would caution that people who think they've diversified are not. A huge portion of the S&P 500 is the top handful of companies, so an index fund may mostly end up heavily invested in those firms.
Some people have shifted investments around to fix that, but I'm sure many people will be quite surprised.
Agreed. If you look at what's in the S&P 500, 7.49% is nVidia, 6.33% is Microsoft, 4.88% is Google, 3.86% is Amazon, and 2.95% is Meta. That's over 25% of the S&P 500 in companies that will see their share prices fall heavily if there's yet another AI winter.
Under diversification, as you are describing, is the first ingredient…
I suspect the second ingredient is US boomers right on the edge of retirement or otherwise shifting to fixed income… And a normal correction causes all of them to divest simultaneously from equities.
There is a potential for a stampede out of US equities.
https://www.slickcharts.com/sp500
NVDIA makes up 7.5% of the S&P500. It will not be a minor blip to 401ks if it collapses.
NVDIA has plenty of other business and will not drop to 0. Even if they drop 75%, than is still only a 2-3% (the math on percentages is weird and I don't feel like doing it - and I shouldn't because a 75% drop is just a random number and so I have no significant figures to work with) drop in my 401k. Of course other companies will be hit as well, but still my 401k will not drop and more than any other stock market crash, and there is every reason to think it will recover. It might stop my plans to retire in the next 10 years (but they were of questionable realism anyway), but any longer than that and odds are it will recover.
Well, that sounds pretty good to me, hope you're right.
I'd say you are underestimating the industries being propped up by the AI Boom.
To name a few examples - Construction (the industry pivoted to DCs) [0], the entire hardware industry (from personal experience CHIPS act recipients pivoted to AI-hype because of slow reimbursement rates), whatever tech hiring that even exists in the US at this point is itself looped into and justified by the AI boom or capital that is 1-2 degrees removed from the AI boom, the entire energy industry is tied to the AI boom [1], and the renewables industry in the US is increasingly AI-washed now as well [2] because of the ending of IRA and IIJA disbursements under Trump.
This wouldn't be a 2008 sized event, but it would be a bloodbath comparable to the Telecom Bust (2001) and the Dot Com Bust (2001). I would trust that every HNer would be worried about such a situation because of how existential a risk it would be for the entire tech industry - which employs most people on this forum.
[0] - https://www.mckinsey.com/featured-insights/themes/whos-fundi...
[1] - https://www.politico.com/news/magazine/2025/08/05/energy-ind...
[2] - https://www.bloomberg.com/news/articles/2025-09-26/ai-boom-w...
I hope it all burns. Zero sympathy whatsoever for AI-anything. It's all a heaping pile of trash.
You hope the garbage heap surrounding your house burns?
[flagged]
How about you start by turning off your own computer?
[flagged]
I’m probably more of a Luddite then you, but it would be nice if it would fail without extending into the rest of the economy.
I would out-luddite you any day don't even try me. I'm posting from a beach where I fly a drone overhead to OCR my stick-writing into HN comments.
You're hoping for a full on market crash? I don't mind AI hype companies crashing, would probably low key enjoy it with popcorn. But from what I understand, it's not that localised.
Yeah it will be funny. I couldn't care less. My tea and tobacco will still be just as cheap as it ever was. I'll still be playing my lute on the porch, unaffected.
You realize the startup you are working on would directly be impacted by any sort kf "AI Winter" right?
My startup is already dead in the water because I built it in an ADHD-fritz and only did it to spite a YC company, and in that regard I succeeded wildly because they threatened to sue me. Haha! I'm working on much cooler shit right now. Sick own though, checking my profile and all that.
It's not supposed to be a burn. My point is being flippant about a potential market crash under the assumption that you can somehow weather it out is extremely unrealistic unless you are a HNWI.
It doesn't have to be bad. Part of the weak jobs market is companies spending their money on GPUs rather than hiring humans. That could reverse.
> Deutsche Bank recently said that the AI hype is the only thing holding the US stock market together.
Doesn't sound right. Russel 2000 is up ~10% YTD.
This 10% is largely due to weaker dollar though.
It’s pretty scary. According to Barron’s, MicroStrategy, a bitcoin treasury company, alone makes up about 5% of the U.S. convertible bond market. That’s remarkable given that it isn’t a typical tech or biotech growth company issuing convertibles, but essentially a Bitcoin treasury company.
The banking sector in the US had assets north of 12 trillion dollars. AI last year had about 100–150 billion dollars of market cap.
I think because we hear about AI so much, we tend to exaggerate its importance?
> The funding ecosystem will be set back years as this wipes out a bunch of VCs and investors
I don’t think this is likely. We’ve seen VCs overextend into social networks, “sharing economy”, B2BSAAS, machine learning, developer tools, video games and crypto, in the last 15 years, and they have very little to show for it. Something new will come along and they’ll invest in that.
VCs typically invest money invested in them, so if that dries up, regardless of what dream they want to chase, they can't. The hypes you list all happened, but I don't think I've seen anything in the last 20 years that even comes close to the current AI hype.
The real question is if enough fails to cause systemic risk. Hopefully the GSIBs ( https://en.wikipedia.org/wiki/List_of_systemically_important... ) aren't too badly exposed.
I think the debt is what makes the risk systemic. Sure it can be overvalued, but alone that won't cause a generationally painful economic meltdown. It's the spending and debt that make something that is a bubble into something truly dangerous. All the exotic private credit and structured finance that is powering this thing along with the lack of any viable revenue stream to keep up with the debt plus interest that make this thing so dangerous. Sure maybe Meta can take a huge hit and limp away (it's still gonna be very painful for them and people who own a lot of its stock) but can OpenAI, can CoreWeave or any of the firms that lent to them? It's a domino effect. The fact that Meta is doing this is a huge red flag. The problem is the market is rewarding this endless cash burn without any way to generate the appropriate revenue. Once reality catches up there are a ton of knock on effects.
> but the most experienced trusted folks I know are already repositioning themselves to weather the upcoming storm.
Repositioning in what way?
The same way they always do every single year: they re balance their portfolio so they are never heavily invested in any one thing.
If you believe in AI and want to bet strongly in it - which some experienced folks do - you take 5% of your portfolio and bet that in AI. The other 95% is invested in a diversified portfolio.
There are many inexperienced investors. Anyone can ride a bubble up and make a lot of money. There is no reason to think you can call the top of a bubble (or if there is a bubble!) consistently enough to bet on it.
And what about in the private markets? YCs entire recent cohort was AI alongside many of the “top” firms.
I'm not 100% sure how YC works, but generally places like that are invested in by people who use it as the risk portion of the portfolio. So you while YC might be 100% in AI, the people behind YC or only 5% in and so it works out. If you have 100% in any company then you are diversified and need to fix your portfolio.
It is generally best for a company to specialize in something they do well. It is possible that YC has picked winners because their specialize in this, and thus even though the AI bubble collapses they are okay. Who knows - but this is something specialists can do in some cases
What you’re describing is my understanding of how VC is designed to work.
I don’t even mean that as cynical— the model is designed to spread risk and let winners emerge, both in company leaders and technology.
That necessarily means periodic culling.
> The startups ecosystem will suffer extensive and catastrophic damage.
These a feature, not a bug of how the startup ecosystem works.
> The funding ecosystem will be set back years as this wipes out a bunch of VCs and investors.
Once again, a feature. There's too many VCs and too many funds.
> the likes of which we haven’t seen since financial firms imploded in 2008.
Institutional financials firms dwarf VC private capital. This alarmist comment makes it sound like a nuke going off in a Nevada desert is going to kill a major US city.
No problem. They can use some of that data center compute power to mine bitcoin.
https://www.whitehouse.gov/presidential-actions/2025/03/esta...
Some of the “AI startups” will be sold in fire sales for parts so investors can at least minimize losses and most of the rest will vaporize
Sometimes literally.
Every time a tech bubble pops, it's an opportunity to upgrade your work-from-home furniture cheap.
I've gotten some nice chairs and even have a nice coffee table that used to grace an office lobby in the old WaMu tower in Seattle.
I guess that makes me a home furnishings vulture.
> this wipes out a bunch of VCs and investors
Hard to see this as a bad thing.
META free cashflow last year : $20bn. Cash on Hand: $47bn. "Worrying" Debt: $15bn sought.
ORCL, the other company they're talking about: $20bn in Cash from Operations, $21bn in capital expenditures, ORCL Cash on Hand: $11bn. ORCL's recent debt flotation: $18bn.
ORCL has 40 years to pay back; demand was reportedly $88bn for the offering. I imagine pricing was close to T-bills.
These flotations posit that demand for compute will continue to increase over the next 40 years, and that infra providers who can get there early will do better than Treasuries.
Calling it a bubble just because the numbers are big is the weakest of financial journalism. Now, do you think inference and datacenter demand will drop? If so, it's worth asking if and when these datacenter will pay, and if they don't, who will take the hit. That would be useful analysis.
I'm pro these plays -- right now inference has an 80% margin; that's after paying the fully capitalized costs of datacenters + compute + the datacenter margin. To the extent a company controls its own inference stack and can do so with a 5% cost of capital, they should do it.
> Calling it a bubble just because the numbers are big is the weakest of financial journalism
Yeah, exactly. The current AI investment wave could be a bubble. Or not. If anyone knows, there are millions to be made in the stock market. The answer depends on the actual expectations for those investments and how actual business metrics are tracking them.
But pointing to ambiguous “evidence” just adds noise. If someone screams fire in a movie theater journalist should report if there’s a fire or at least smoke, not write articles showing concern about how flammable all the furniture is
The thing is, the models do work. They add value to me each and every day, to a degree almost no other tech has done before.
But that doesn't take away the fact that this is extremely expensive stuff (not for me, but for the companies pushing the envelope), far too expensive. And it is really taking its toll on other resources, like electricity.
I think the thing that most of the folks miss is exactly that. AI is adding value and it’s here to stay. Absolutely the OpenAI story is looking concerning but I am not sure it is this doomsday AI winter.
Both can be true at the same time. Similar to the early days of the Internet, the dot-com bubble eventually popped, but the Internet (and dot-coms, for that matter) didn't go away.
What people are saying is that this mad race to throw cash at anything that has "AI" in it will eventually stop, and what will remain are the useful things.
No the tone is generally that some mythic AI winter is going to happen because of current valuations and that AI is simply the current crypto grift.
Yeah, I saw "AI winter" mentioned elsewhere in the thread...
IMO there is a real qualitative difference between AI and crypto in terms of the durable impact it's going to have on the world. Does that mean I've bought into the AI hype? Maybe. But I think the signs are there.
AI winters are a recurring phenomenon, not a myth, and, like the dotcom bust, involve a collapsing hype bubble, reductions in focussed speculative investment in the field, but the technologies that were big during the preceding hype cycle continuing to be important, and develop, though in the case of AI winters often they stop being thought of as AI and just get referred to with a name for the specific technology (often a different one than the main one they were known by in the hype cycle, e.g. “expert systems” from th blast hype cycle are largely “business rules engines” now.)
Googling, there seem to have been two AI winters, the first (late 1970s - early 1980s) when people first figured AI was overhyped, and the second (late 1980s - early 1990s) with the collapse of expert systems. I don't think we are about to get one now - more like AI spring leading to AGI summer.
If revenue doesn't catch up with the cost of developing and running these huge LLMs then the only way to avoid an AI winter is to find a way to make them way cheaper to develop.
Inference costs are largely break even and profitable. Now yes there is absolutely questions on the training side.
You could turn off the switch today though and most of your leading models could keep the lights on.
I keep seeing everyone guessing what the margin is on inference. You say that it's largely break even. We have this person in the thread claim 80% margin (https://news.ycombinator.com/item?id=45462442).
These numbers are complete make believe.
Why would they be make believe? I am generalizing across multiple companies and simplifying it as much as possible as we won’t know all the cost buckets but we do know at current costs the bare inference cost of keeping the machines running is being covered.
But are you paying for them commensurate with the value they create? That's the problem. Tons of money being invested/borrowed on technology with not nearly enough revenue to justify it. Same thing that happened in the late 1990s.
Actually the problem is the same as all previous technology: is the company getting 100% benefit while employees get less than zero with lost jobs, lost negotiating power, lost unique skill marketability, etc
Of course not, that’s what make vibe coding “fun”.
A bubble doesn't imply the product is bad. We all need homes, but there have definitely been housing bubbles.
No question that neural networks can add value, as has been true for a long time. The question is, how much? The current expenditures indicate that the answer is, "trillions of dollars worth".
If you call it "neural networks" or "large language models" or "machine learning", this would sound absurd. Only by calling it "AI" (with the implication that it will achieve general intelligence) does this sound, for most people, possibly correct.
Add what value though, enable you to write more slop to power more slop that requires more slop?
https://archive.ph/rou3t
The entire US stock market and economy is being prompted by AI since 2022.
I think that the Economist posted some months ago a study that indicated that without datacenter capex booming the US economy would otherwise be flat.
At some point there will be major shakeups.
It's just gotten so out of hand that there are multiple levels of pain. There's all the debt financing / private credit deals on infrastructure costs that will be defaulted on. There are the losses that most people will feel when the overvalued shares drop. There is the pain of tech companies having spent so many resources with very little in terms of viable revenue streams. It's gonna hit a lot of places in the economy.
> "SPVs mean companies like Meta do not need to show the debt as their debt," Perkins writes in a note. He likens today's financing tactics to the subprime era when firms shifted risk off the books to reassure investors.
But it's not? If the SPV is doing its job as a limited liability entity, Meta should be bankruptcy remote from the SPV. Presumably accounting rules would look through that SPV if Meta was actually reachable. It doesn't make a good comparison to GFC where everyone was nominally on the hook but had that risk "insured" with CDSs, etc.
Perkins’ point is valid. It’s still shareholder assets and resources at the end of the day even if via a SPV. The banks did this sort of thing in the financial crisis with all sorts of subsidiaries and circular funding.
“Hey it’s OK that mortgage side business is going belly up but we bought that special insurance.”
“Cool, so who sells that insurance?”
“Um, actually I think we do via our other SPV”
“$&@!”
Not 100% analogy here but similar shenanigans where company A is taking out debt, to build a datacenter, to buy GPUs, to sell services to startup B that Company A gave cash to to buy said services. Startup B has no viable business model and implodes. Company A goes from looking like a genius to a financial nuclear waste dump almost overnight. It’s financial shell games all over again.
Isn't the problem the creditors exposed to the SPVs though?
Meta, and others might be shielded from it but there's someone on the other end of these debts, if the SPVs are accruing massive amounts of debt and go bankrupt there are lots of obligations that will go unfulfilled, bringing the whole house down.
It's quacking very similar to other accounting tricks, caham, financial engineering, to hide bad numbers somewhere else. CDSs and CDOs were a different mechanism with the same purpose: hide risk away into overly complex instruments.
> "SPVs mean companies like Meta do not need to show the debt as their debt," Perkins writes in a note. He likens today's financing tactics to the subprime era when firms shifted risk off the books to reassure investors.
The article doesn't explain how Meta are actually using SPVs. The suggestion is that the debt is in fact Meta's debt but the use of an SPV somehow means they don't need to show it on their balance sheet. Traditionally if you're using an SPV it genuinely means it's not your debt, in that you won't go bankrupt if the deal flops and the assets are underwater. (And they're not your assets, either.)
There are valid concerns about debt fuelling outsized investments in AI, but this article isn't really saying anything of substance. Plenty of non-bubbles attract significant debt investment as well.
Ultimately, debt is a way to transfer risk. It becomes an issue if it results in a lot of risk being transferred to parts of the economy that can't really afford it (like retail investors, or systemically important banks). As long as the people taking on the risk know what they're getting into and we won't need to bail them out if they lose their shirt, it's not really a problem.
Is there a bubble? Maybe.
Will all of the current companies exist 5 years from now? Probably not.
Is AI generating value and here to stay. Absolutely.
Can "AI" make me a React component that doesn't blow up in my fucking face? Absolutely not.
That itself is not proof that AI is not here to stay. It’s a short minded world view. Value is absolutely being generated but of course not everywhere and I am certainly not defending valuations.
Value is being generated for owners. Not for me. I don't care at all about "value". I don't know what perspective you are thinking from. Are you rich? AI and crypto and every flavor of the year tech garbage are just siphons. Judging the tech on its own merits is missing the forest for the trees.
Dismissing an entire technology because you personally haven’t figured out how to make it useful is a cope, not an argument.
Am I rich? What are you even on about.
I can now spend a few dollars and classify a large dataset that would have taken me real time before. I can again spend a few dollars and turn unstructured data into structure. There absolutely is value to be found in the current gen of tech. Comparing it to crypto is a weak argument. Just because you don’t find value does not mean other organizations and people are not.
>Am I rich? What are you even on about.
The only people I find harping on about some amorphous "value" floating around in the air are typically people with portfolios.
>I can again spend a few dollars and turn unstructured data into structure.
Saving 5 seconds on a 10 second task using an infinite amount more energy is not "value", it's garbage propped up by VC money and regular people are going to be paying for it when it inevitably goes tits up. Like we've paid for everything. There's your "value".
You’re confusing your frustration with some half-baked React scaffolding with the entire field. “Saving 5 seconds on a 10 second task” is just a strawman, there are workloads where it saves weeks, even months. If you can’t see that, fine, but don’t pretend your blind spot is universal truth.
And spare me the VC boogeyman. The fact that capital flows into bad bets doesn’t erase the actual wins. We’ve had garbage propped up before, railroads, dot-coms, mobile apps but guess what? Railroads exist, the internet exists, mobile exists. Same pattern here.
To save months the output would have to be reliable, but it isn't. It saves very little, especially at data input and coding.
It does save time for tasks where the output is easily checked, such as image generation and translations. But the quality is often mediocre.
I cannot say for your work but for classification steps and data structuring it’s quite accurate and this is with regular testing. I cannot speak for your work but for mine and folks in adjacent industries, LLM are fantastic and adding a lot of value to our workflows. You’re honestly holding on to this dead idea that LLM outputs are full of hallucinations. Throwaway account with throwaway comment.
It's "quite accurate" which is not acceptable for almost all relevant tasks is a business context. Somebody needs to manually check everything. Almost no time is saved.
Talking as someone who has built many small OpenAI integrations aka wrappers in business apps.
I would pay good money to be this green again. I talk about regular people, and you revert back to spreadsheet efficiency. Classic tech guy response, just cannot see a bigger picture.
“I’d pay good money to be this green again” just reads as bitterness dressed up as wisdom. You’ve offered zero substance beyond sneering at anyone who finds use in the tools. If your whole argument is “I don’t like it so it’s worthless,” that’s not insight, it’s just cope.
Not sure starting to appear is the right word
Coreweave had a huge chunk of debt and gpu financing was a thing even last year
When you look at total real returns over longer time scales, the current period in time is not unusual.
https://totalrealreturns.com/s/SPY
We're not far from the long term trend of 6.15%/yr on the S&P500, just slightly above.
The deviation above the long-term trend was much larger in the run-up to 2001. It's like we're in 1997 now.
Not saying a correction isn't likely, just that it's easy to scare yourself by looking at linear, non-inflation adjusted plots.
- ChatGPT still lags for me daily and I'm on a PRO subscription. - SORA is usable but extremely slow and constrained. - Image gen same thing, but getting a bit quicker and less constrained.
This is with agents not having taken off yet, and the vast portion of the world economy not interacting with LLM Ai? Agents alone will require ungodly amounts of compute.
Bullish
> ChatGPT still lags for me daily
What exactly are you referring to? Their UI seems to have troubles to load the list of conversations since like a week back, otherwise it seems fine? I mean, UX could be a lot better, but I'm experiencing anything I'd describe as "lag".
Outputting code mostly, especially compared to Gemini and Grok Fast. I'm using normal thinking mode, not extended or PRO.
>list of conversations
How many bytes could that entire list possibly be? Surely less than your average webpage ad.
Yeah, especially considering it's only fetching ~10 items at a time :) I'm guessing some index somewhere is wrong or something, as I probably have a couple of conversations through the years of using it almost daily...
This will have a big effect on those in passive index funds I think?
Tech made up about a third of the S&P last time I checked, but the AI hype is also affecting non-tech sectors.
Dang. My kids are just getting the hang of using ChatGPT as a way to generate random quizzes and additional homework problems; to the point where both of them just automatically use the AI to "fill in the gaps" each evening. (Prior to AI, additional HW was met with less enthusiasm than you'd think!) I told my wife that the billionaire class spent the last few decades cheating the tax system (and, thus, the education system) just to weirdly dump a trillion dollars into the AI ecosystem to build the world's finest tutors. I really hope that doesn't Go Away™.
It absolutely won't go away.
Even if OpenAI folds, the open source stuff is good enough that someone will build a compelling tutor platform in their place. It probably already exists.
Don't worry about this. There's no way these tools are going away. If this bubble bursts it may wipe out the incentive to continue this frenzied race to build novel AI, but ChatGPT et al won't be shut down, and even if they were, the open-source LLMs comparable to the cutting-edge of 6 months ago will still be online and available. Plus, even if AI progress froze solid tomorrow, I think it would take decades before we'd start to anywhere-near-saturate the potential space of applications & use cases to really do the current tech level justice. (Also, even post-bubble, AI progress would not freeze solid, to put it mildly)
Are you paying anything for it?
>Meta seeks $29 billion via private capital for its AI data center buildout.
This would be worrisome if they didn't have the cash to pay for it, but that's peanuts to Meta. It seems more like tax avoidance schemes than anything else.
First Brands and Tricolor Are Signs of What's to Come: https://www.bloomberg.com/opinion/articles/2025-09-29/first-...
Archive: https://archive.is/rwJuP
$1.6T Credit Bubble Bursts as Two “Healthy” U.S. Firms Collapse: https://www.youtube.com/watch?v=51DmcBx6BI0
The bubble will burst with a vengeance when big corporations and startups start deploying and utilizing LLM models on-premises instead of the "pay-as-you-go" trap.
B... bb... bbbut i thought we are so close to AGI and all the AI companies can easily make bazillions off of targeted advertising and and AI can do all my employees jobs for 1/1000th their salary and and...
Hey now! I can’t raise money telling VCs the truth!
The job market is already a joke. 2026 is going to look very bleak.
The only people talking all the time about AI bubble are those on HN. I seriously wonder how this place got so huge in the valley.