Big money is nervous about AI hype, but not ready to call it a bubble In an employee share sell-off this week, OpenAI achieved a nominal value of $500 billion. In terms of valuation, the posterchild of GenAI — which is yet to make a profit — left in its dust companies like Toyota, the world's largest automaker. Non-believers might be interested in whether this combination of escalating asset …
"not yet sounding the alarm, but was careful to add a note of caution."
Such financial commentators are probably realising they're treading a tightrope.by now. An over-strong remark could start the whole thing collapsing and they won't want to be named as the one responsible. On the other hand they could risk getting sued with recommending something they should have known was going to collapse causing some punter to get burnt.
For those with the right tools, it is possible to track how often expressions referring to an investment bubble in <topic> occur in news articles each year. Whilst most people only recognise a bubble in retrospect, there is always press chatter in advance that rises in volume even if it isn't a mainstream narrative. I'd be very interested to see the results for searches against the dot com bubble and GFC, and for the latest AI bubble - and for dot come and GFC we have investment indices to compare.
"LLMs can confidently give an answer that is completely wrong."
Unfortunately that's a highly valued skill. Look at at all the crap pumped out by consultancies, think tanks and so forth - often perfectly polished, attached to a widely known "brand", and then swallowed by business execs as though it's gospel truth, usually in return for a huge bill. I used to work for a major energy business, and nothing happened unless some top consultancy were involved as advisors. BCG, Accunture, McKinsey, Bain et all were busy filling their boots charging top dollar for nothing new, but presented with confidence. Couldn't even do a simple restructure to get rid of 100 roles without having some consultancy wankers dress it up in fancy slides and get to speak to our board (us proles weren't admitted, even though we were identifying the new structure, and implementing the change), and of course it was presented as something new and fancy, a "Target Operating Model".
Which goes down best: Crap content accompanied by beautiful powerpoint slides and presented slickly by a smartly dressed outsider, or important, insightful content that has drab presentation and is given by a dull-voiced employee in their usual work garb?
And of course, there's politicians who universally speak with confidence and are universally untalented and wrong.
Analysts, AI diviners etc ?
“Your money is at risk and you may get back less than you invested”.
If these scum - the same sort that sided with Putin over Energy, Grain, Sunflower Oil, Fertiliser etc - commodity speculation/Ukraine war want better protections than Joe Public …. They can fuck off.
I would say that this time it appears all the usual finanical news channels I keep on tap are almost all asking "is it about pop?", so most are very aware that AI companies are most certainly overvalued and it's simply a matter of "WHEN" you jump not "IF". Just be sure to be well clear of any involvement when it does pop!
And about as useful. Just need the equivalent of Dr Pimple Popper to make it go away
The longer it will take to "pop", the louder the bang will be...
The AI bubble is different in one respect from earlier economic bubbles, namely that the path companies go to try to make AI at least issue somewhat "correct" answers seems just to be: "make it bigger" for the past several years.
This leads to the need of accelerated spending across the AI supply chain (accelerators and data centers, both age fast and require short amortization time frames, lots of energy), forcing the industry into an exponential financial (and environmental) trajectory - to the "pop".
Just try to notice how often you nowadays hear the word "trillion" from companies yet to make _any_ profit trying to make the already sunk cost look reasonable.
The longer this sunk cost fallacy is maintained, the harder the crash will become.
I'd be surprised, if there would not be a major correction before Q2/2026.
Ladies and Gentlemen, fasten your seat belts, please.
I don't think the big tech companies are stupid enough to invest their own money on some breathless marketing nonsense. They will have taken insurance and even insured that insurance. The people who will pay the bill at the end will be the little investors who thought they bought a piece of a license to print money.
In gold rushes the ones selling shovels get rich, everyone else loses.
> It's getting to the scale of worrying what it takes with it
A national state? That has brought into hype and so has leveraged taxpayers wealth to build “AI datacentres”. Naturally the contracts will be water tight, so monies will have been paid, leaving behind a bunch of asset less shells and half finished datacentres…
somewhat "correct" answers seems just to be: "make it bigger" for the past several years.
But with such rapidly diminishing returns of performance vs size, and the really obvious elephant in the room- there have been no practical improvements of this technology since what has always been simply an intellectual curiosity escaped the labs a couple of years ago... It doesn't take a genius to figure out that spending trillions on 0.5% performance increase in some random benchmark resulting in an AI that "now only gets shit wrong 45% of the time" is not a great investment.
Seat belts? I'll take my chances and jump out the door at this point.
The problem with A.I. is that its promise is endless, but its actual usefulness is, beyond generating images and making summaries, very limited and therefore not worthy of the over-hyped speculation on the stock market.
Note that this could change almost overnight if some breakthrough is made, but for now we don't have a clue if this is on the horizon.
reminds me of the dot.com or dot bomb bubble in the early 2000's, all went crazy with over-evaluations of companies. I was working for an online news company which started with 4M VC in 1998 and it went to a valuation of 200m with no revenue, companies just going to get online to get higher stock valuations from the market makers etc. What happened in the crash was probably more of a correction than a crash but it did mean alot of the free stuff started to become subscription services. There were lots of free cloud drives around and this all changed to tiered charging.
High interest rates at the time and less VC available also didn't help with the crash.
We might see the same with AI subscriptions going much higher from the ones that manage to survive. ?
> We might see the same with AI subscriptions going much higher from the ones that manage to survive. ?
Surviving AI will (need to) be much more expensive. At the current rates, the AI operators lose money on each customer.
So AI will become expensive - maybe too expensive to replace all the jobs it currently promises (or threatens, depending on your wealth) to take over.
AI deployments for job replacement in the real world already fail to pan out in the majority of cases, so AI becoming more expensive will further reduce its adoption, further degrading AI company profitablity.
If that turns out to be the case, it's vicious-circle time for AI vendors. I think I'm fine with that.
Prices will rise to meet confabulated demand.
However, there is something unpleasant for major 'AI' producers lurking in the global woodpile. At present, enormous sums are spent to 'train' supposedly 'all singing and all dancing' ersatz 'intelligences'. Current LLMs do exhibit fascinating properties and undoubtedly have a variety of practical uses. Not to be discounted is the arrival of a parallel not-for-profit market arising from independent institutions, including groups of academics and engineers.
LLMs have sparked interest, but it shouldn't be assumed that software entities of their nature, and more advanced, cannot arise from simpler templates. Moreover, energy requirements for training and use may not, in principle, be gargantuan: the biological role models for current AIs, by comparison, are parsimonious users of calories.
Independent software engineering may focus upon specialised AIs for specific tasks. Training schedules need not attempt to encompass every digital artefact ever devised. If one wants aids for specific professions (e.g. lawyers), particular tasks (e.g. supervision of automated production lines), or for general use as curators and access points to genres of knowledge, their construction doesn't require either the size of model or the use of electricity associated with LLMs indiscriminately fed from everything representable in digital format including the output of morons abounding on Twatter.
Also, noteworthy is the ability of 'consumer' devices to run trimmed AI models in situ. Personal computers with suitable graphics cards can do amazing things. Furthermore, pre-existing models can be 'refined' or externally adjusted (e.g. with LoRas) to better suit particular users.
Additionally, models fitting within run-of-the-mill devices rather than supercomputers will, as is so for other digital materials, be freely shared among people wanting them. Eventually, there could be an abundance of AI models arising from around the globe and competing for attention by virtue of their knowledge-bases, functionality, ideological underpinning, degree of censorship, and freedom from propaganda.
Large amounts of money have already gone into it and people are queuing up to shove in more. Unless the ventures start to turn profits commensurate with all that money it will be lost. We're already at a point where the required returns appear infeasible without it enabling some huge economic advance that's within the capabilities of the planet's resources to sustain. Historically technological advances accompanies by investment booms have stimulated economic growth but (a) investors have still lost money (George Hudson's investors in railways for example) (b) it remains to be seen whether economic growth on the scale required is sustainable. (c) It remains to be seen if there are any economic advances at all to come out of this.
I'm not sure, to be honest. I feel like open source models that are "good enough", and can be run locally, might be what will be left. And Moore's law is not entirely dead yet, just moribund, so they will soon reach a level around about as useful or useless as the best current models.
Having also been of age during the dotcom crash and the 2008 crash, I don't think either were as grotesque as the current AI Ponzi scheme. The amount of swamp gas inflating it is exponentially higher than those two bubbles. It's just waiting for the spark, and as James Burke said "BOOM".
I just wish I were smart enough to know where a safe haven for savings lies. I suspect there isn't one.
Let's unpick this drivel:
"It's going to be in every TV."
Yeah, OK, I can see how *GEN*Ai might be helpful to suggest what I watch next, but we already have that. Hardly cutting edge. Now how about it Generates a new episode of something I really like, on the fly - now that *WOULD* be a cool addition.
"It's going to be in every phone. "
Why do I need *GEN*Ai in a phone? My current fruity phone already indexes my photos to allow me to search for *that* cool pic of the dog last year. We are already at peak Ai in the hand set No?
"It's going to be in your car."
Really? GenAi in the car? - It looks like you are on a twisty road, other cars that were on twisty roads had to stop suddenly for deer, I'll slow down... Oh, wait, that's not working because I'm in the highlands with no 4g... back to hand steering the car for me.
In your toaster.
100,000 others liked their toast do a bit darker than your old setting of 3.4 out of 5, so I have burnt your toast.
"And in every streaming service."
As above for TV, I already get suggestions.
"Every piece of technology "
Just because it has a CPU, it doesn't need GenAi.
"...that you consume today will have GenAI in it," John-David Lovelock, Distinguished VP Analyst at Gartner, told The Register last month."
...
...
...
We're doomed.
Because if "AI" is in every phone, car, fridge blah blah blah, it just means people like me will not be buying any of those things. So in addition to the pop from AI going bust, add the revenue lost from all the consumer paperweights built that have the builtin cancer. So even the car co's, phone makers, appliance makers are going to see a hit from building product people don't want.
"Distinguished VP Analyst "
So that's what team leaders get called at Gartner these days.
Next year it'll be Distinguished VP Analyst and Demi-God. The year after Distinguished VP Analyst, Omniscient, and Demi-God. By 2030 his job title will be on a superlative laden extra long fold out business card, maybe even ranking up to a whole God.
> "It's going to be in every TV, it's going to be in every [everything]"
Nah, this is just the core argument of any AI evangelist. "Best thing since sliced bread! Tablets Blockchains Clouds AIs will be everywhere, don't be a loser, invest now, invest big!".
We have heard that exact same blurb again and again, just the subject changes.
Have it in my phone than my TV. I don't want any smarts in my TV at all, and don't use it for streaming I leave that to a set top. That way I can insure my "TV" experience is always up to date since I can replace the set top if it stops getting updated but I don't want to replace my TV for that reason. Having "AI" in a TV just makes it cost more for something I will NEVER use, and will be used to fool average consumers into thinking the TV they have now is hopelessly out of date because it lacks AI.
Yes automatically organizing photos and such is nice, but I would love it if my phone became a proper personal AI assistant. Whether/when technology allows this who knows, but I'd love to be able to assign it tasks and it opens and drives the apps to accomplish what I want. Maybe I tell it "keep an eye on my checking account balance and expected bills based on my credit card usage and known upcoming expenses like property tax payments, and when it looks like it'll drop below $2000 initiate a transfer from my high yield savings to top it up" and it opens the apps as necessary to keep an eye on all that and does the transfer. I wouldn't trust an AI to do that today, but eventually maybe I do. I definitely want such an AI to run directly on my phone, not only for privacy but because that's the only way it is going to be able to gather all the info it needs and execute my wishes via multiple apps.
Basically I'd want it to relieve me of as many of the "background tasks" one needs to perform to function as an adult. It can't go to the grocery store or do my laundry, but anything that has been automated via technology accessible from my phone would be fair game. I wouldn't set a reminder on my phone for a week before property taxes are due to pay them, then move that reminder to 6 months from now. I would tell my AI once and then never have to think about it again.
Except an LLM cannot ever do any of the above, because they're text generators.
Rules like "use this savings account to ensure I do not go overdrawn" can be trivially implemented as an actual testable algorithm. I am certain many trading systems actually do this already, except they're proprietary and not available to consumers.
You're right of course, but unfortunately most people aren't capable of writing a testable algorithm - or at the very least are so unwilling to learn how to do so that it equates to the same thing. "AI" promises to make things that require a bit of thought and care "Easy". This is of course a lie, because if you're not sufficiently specific with what you tell your "AI" to do, it will probably do something undesirable - and even if you are specific with telling your "AI" what to do there's a good chance it'll do it wrong.
Problem is that it's a really attractive lie.
Don't want to learn how to use your banking app? Just ask the AI to do it for you! Don't want to learn how to program your VCR*? It's fine, just tell the AI what you want to record. Don't want to learn how to drive? Get a self-driving car with AI in it to drive for you!
Obviously some of these examples are easier for machines to do successfully than others - but the lie is the same lie in each case. There's a job you don't want to do, so here's a friendly robot that can do it for you, and all you have to do is tell it what you want.
"Take me to Stockport" is an apparently simple instruction that most taxi drivers are going to have no problem with, but being human they're likely to point out that Stockport is quite a big place and you probably want to go somewhere a bit more specific. If you said it to a taxi driver in Solihull they may well be happy to pick up the fee, but again I imagine most of them would first double check that you actually mean it because it's better part of a 2 hour drive and are you really sure you want that on the meter?
Most apparently simple interactions are a bit more complex than one would immediately assume. To provide complete instructions to make a cup of tea, one must first invent the universe etc. "AI" doesn't do well with that and it's not clear that it ever will, but the promise of AI always glosses over that part because it interferes with the very saleable idea that the nice robot can do all that difficult thinking for you.
*Let's pretend those still exist
My rather old TV has an AI upscaler. When it shows an ordinary broadcast resolution source the AI adds details. When someone moves across a background the parts of the background about to be covered and recently uncovered have a twisting flicker that I find irritating. My older dumb TV does not do that.
The current state of AI is irritating with no off switch. The only advance I am expecting is that the off switch will require a subscription.
What you describe sounds more like it's increasing the frame rate by trying to predict tween frames. My friends 4K 60Hz samsung does this if I feed it a video file encoded at 24,25 or 30Hz. If I feed it a 25 or 30 frame video file encoded at twice the rate, such that each frame is duplicated, it doesn't fuck with the image and displays what was in the video file. The problem is that there is insufficient temporal resolution at 25 or 30 Hz to be able to make accurate estimation of motion, and the process is completely unable to understand movement of light and shade. It appears to be using the motion information directly from the video file, which doesn't account for occlusion, which is probably the source of these artifacts. A lot of people seem to be blind to this, but i find it rather jarring when bits of the image that my eyes are naturally drawn to drop the frame rate.
Neural radiance fields may one day solve these issues, and allow you to move the camera position too.
Someone needs to work out the consequences when the bubble pops, AI goes TU, and this becomes an extinction event for those who were all in on it.
AI has no ROI and is a major security risk. In use, it is generally unreliable, needs as much human oversight as a 2 year old toddler, and is consequently no great benefit in the workplace.
Trump is squeezing investment out of his allies ($500bn from Japan) which would outsource much of the damage to other countries, if they pay up/underwrite investments before it goes pop.
The capitalist economy is not based upon people being forced to pay for things they don't want and have no use for. And there are enough of us pointing out that the emperor has no clothes on.
Some aspects of AI will find niche use amongst the wreckage, but a hypothetical crash model would be an interesting read.
People sitting atop bubbles with a parachute to hand descend unscathed from market-crashes.
Referred to are operators clustering within market centres such as Wall Street and the City of London. Some have secure incomes from acting as middlemen between buyers and sellers of stock. These have incentive to talk-up values when their transaction fees are percentage based. Others working under fixed fees benefit too by encouraging a frenzy of buying.
Financiers sitting on piles of a stock with a view to a huge capital gain do bear some risk, but it needs only be minor. When holding stocks which have doubled or tripled in nominal value, they may anticipate a small paper loss at the outset of a bursting bubble or broader market crash. Unlike most private investors and managers of small pension funds, operators at a financial nexus benefit from three things.
First, active participation in the flow of information concerning 'insider' knowledge. These people recognise tremors indicative of seismic shock to come.
Second, algorithmic trading enables rapid pre-prepared responses to tremors. Physical location close to a trading centre/exchange offers an edge over location further away; a simple consequence of transmission times along cables. Better yet, to have a direct connection to an exchange to cutout circuitous routes via ISPs.
Third, exclusion of almost all private investors and their advisors from timely public information about stock prices. Without subscription to an exchange, information flow is deliberately retarded by fifteen minutes in the UK and similarly elsewhere.
Thereby, markets are rigged. Big players respond immediately to tremors. If prices have gone down, but settled, they can buy back-in advantageously. Should a bubble be bursting, they will bear some loss on the previous notional value of their stock, but they will have factored that in as something to anticipate and to weather.
Few people raise complaint about the two obviously remediable blights of algorithmic trading and imposed delays on information passage to 'outsiders'. Each is a preventable scandal. Computer assisted trading is here to stay; however, it should have built-in, unavoidable, delays emulating those when trading intermediaries thronged exchange trading floors. Delays reduce the risk of out of control feedback loops occurring as algorithm speaks unto algorithm. Analogously, social media platforms such as Twatter would be much less prone to inane 'viral' episodes if the mechanism of making responses contained unavoidable delays. As an aside, it's more likely that social media operators would offer subscription payers shorter delays so that their responses gain immediate prominence.
The advent of the Internet makes it inexcusable for exchanges to delay output of the activities for most people. Privately owned, exchanges may be, but they are public resources. Real time information can be entered onto the Internet at negligible cost.
Unfortunately, banking and finance, especially since "deregulation" are laws unto themselves (as indeed is the City of London literally). Of course, the true underlying rot - fractional reserve banking - spans centuries.
If/when AI mania collapses, one may be sure of 'little people', via private investment or through pension funds, bearing the consequences.
Oracle "says it has a $455 billion spending pipeline from its AI datacenter customers."
I wonder if that demand is similar to the 2 million pre-orders for Musk's Cybertruck? (which evaporated)
Having spent the last 30 years working in the Oracle space, this is like so many other promises... just hot air.
i bet $300 billion of that is the bullshit openai deal that openai can't afford.
a DC costs $15 billion to build EMPTY, you're looking at another $60 odd billion for the compute. And they want...20?
And at the end of it all....." who could have seen this happening?!", "this was totally unexpected!", "a once in a generation fuck up".... except at 50 this is my 5th or 6th "once in a lifetime" fuck ups
Point 1. The main players in the dot.com bubble had potentially very profitable services. Amazon had a very obvious and useful business model selling books I couldn’t get in the local bookshop over the internet.
Point 2. AI does not really exist. The successful applications are really just better pattern recognition or search plus cut and paste. I would be more interested if they labelled the products correctly.
Point 3. When in a few years time they get this stuff to work there will be some “good enough” open source software so profits will be non existent.
Working people again. As always, the powerful say: "You're ugly. You're disgusting. I'm gonna kill you. Give me $200" and they not only comply but hum along, hoping this time, their rich friends and politicians will like them enough to let them into the club.
I wish I was able to move up the ladder and stop being a sucker but I'm not cold-blooded enough to do what needs to be done.
What was the saying, markets can remain irrational longer than you can remain solvent.
However this does not apply when the amounts of fresh capital needed every year to keep inflating the bubble reach biblical levels. At that point, gravity must take hold. Nvidia and Oracle's circular deals indicate this is quite near now.
generative AI does not exist, it's a giant hallucination machine. it can't do anything other than hallucinate.
ive always hated the shysters at Gartner & annoyingly in the media space it's only Ed Zitron that seems to be saying the bleeding obvious.OpenAI ALONE needs $trillion pay its bills and cover the promises it's made. For a product NO ONE wants. Ove literally just removed Office 365 from my PC because of that copilot shite.
The fact that none of these idiots can see an issue with Toyota being worth 1/2 openai is stunningly depressing. I mean i would love to get paid to spout bollocks like that. Cars make profit, openai loses on every prompt.
An example of this bollocks is the claude code league online to see who can rack up the biggest bill on their $200/month subscription. the leader is some dude whose run up $50,000 cost in a month after paying $200!! for a product that doesn't work!
if you thought working out cloud costs was impossible, how many tokens are YOU using as an enterprise & how can you predict it over the next 6 months?
there literally isn't enough VC cash to carry on funding this nonsense
AI is not intelligent, it is just search:
The bubble will burst, the promises of billions being spent will simply evaporate, idiots who went all in will lose their cash, the con artists at GAFA will go back to the drawing board to work on their next big scam, the media will find a new moral panic to push to the proles, any niche uses for AI will be quietly rolled out rebranded as ML, and nobody will mention 'AI' for a bit, like the Metaverse, NFTs and Brexit.
Moral: Do your own due diligence. Don't follow the herd to the abattoir.
which is fine if you're doing your own investing on the side. my pensions are heavily invested in the S&P asx well as the tech firms. I have zero control of that unless i want to take up looking after my pension full time. hopefully these large funds will sell off before the normal investor gets the news but still... i can genuinely see my pension disappearing
The U.S. stock market is one big bubble and the government debt and budget deficit are the mother of all bubbles.
At some unspecified time in the future the music will stop playing and the Western world will be left holding the bag. China too will be adversely affected but probably less so, since it produces actual goods that people need instead of vaporware and overblown stock valuations.
Ed Zitron has a good article on AI bubble financing, but this pic is the essential part of it:
https://www.wheresyoured.at/content/images/size/w1000/2025/09/data-src-image-ed810554-1dd1-46fc-aecc-9cbd75eb5073.png
Everyone is giving everyone else hundreds of billions and that money actually does not exist.
Those are all ways to look at it. Another is possibly even more worrying. It's boring everyone to sleep , err, sorry doing accounting. Which is what The Economist was talking about two weeks ago.
Basically, after spending huge amount$ on nVidia's latest and greatest, what rate will that latest and greatest depreciate at? These are bleeding edge chips, in a bleeding edge froth of innovation. They are not where say the X86 server chips are currently at in terms of expected durability.
If they depreciate over 5-6 years, that means X profitability for its buyers. If it's 2-3 years that profitability drops. Massively.
AI is a fantasy for owners and managers, who love the idea of a mathe-magical process that can further weaken the wage worker's negotiating position, siphon the remaining money to their pockets and enable security forces to more efficiently surveil them when they get uppity ideas.
As such it will continue to be a black hole until that illusion is shattered.
Dammit, but LLM sure looks plausible. Of course, LLM can never be “artificial general intelligence,” but people won’t believe it unless it’s proven definitively. So valuations persist.
(LLMs never had any interaction with the real world, so obviously everything they state is a “hallucination.” LLMs cannot lie or exaggerate, except if they’re trained on false or exaggerated content. LLMs are incapable of thought, is there any question? As it so happens: yes, some people believe in LLM! Bewilderingly, some people believe in that crappa junk.)
I've decided there's a gap in the market for someone who can confidently make shit up and I'm going for it.
At the moment I'm more expensive than Real Artificial Intelligence but, as the GenAI companies try to pay off their investors, I'm confident I will become competitive.
I don't require much power to work and my only backend expense are substances to cause hallucinations, boost my confidence and damage my memory.
