AI's $344B 'language model' bet looks fragile

https://www.bloomberg.com/opinion/articles/2025-09-11/ai-s-344-billion-language-model-bet-looks-fragile

by thm • 10 months ago

91 117 10 months ago

redwood3• 10 months ago

Oracle's mega moves in the market (cpl hundred $B in market cap),due to their claim that OpenAI was doing a multi-year commit to move much of their workloads to their cloud ... likely a heavy revenue play rather than profit... with an aspiration to push as much CapEx deprecation to out years as possible (btw Oracle where is all the capex?) AKA Financial aengineering... just shows how overly leveraged this bubble has become.

The Economist recently featured a piece pointing out that it's no longer risk that drives the market but a balance of fear of loss and fear of missing out (https://www.economist.com/finance-and-economics/2025/08/06/w...). FOMO is out of control right now

jakeinspace• 10 months ago

This Oracle surge and revenue predictions really feels like jumping the shark. I mean, it's Oracle.... I've never felt confident enough to bet against a company, but a short position on Oracle may well be too tempting for me.

brookst7• 10 months ago

I lost money betting against Oracle. It taught me to never underestimate the power of a sales and marketing organization, even if the product they’re selling is technically backwards.

andsoitis• 10 months ago

You probably saw in the news yesterday that Oracle stock shot up because they are scoring large AI deals: https://www.wsj.com/business/earnings/oracle-stock-orcl-ai-d...

fidotron• 10 months ago

FOMO is out of control right now

Exactly. The whole stock market is currently behaving like the crypto bubbles.

amelius• 10 months ago

https://archive.is/2rFK4

redwood3• 10 months ago

This technology demos incredibly well and you can just see how everyone gets giddy with excitement around using it. I watch my colleagues and executives proud to show what they could do or make endless jokes about it. It reminds me of when people first got their phones and couldn't stop showing everyone how cool they were.

This leads to an over rotation in the perceived value.. the value is significant just as the mobile phone was, but not going to live up to the hype in the near term.

It's definitely interesting how in anonymous forums there's a lot more people pointing out that they think this is hype whereas when we wear our professional hats many of us join in. It's like we all want you to party going no we all know what's going to happen

wffurr3• 10 months ago

> in anonymous forums there's a lot more people pointing out that they think this is hype whereas when we wear our professional hats many of us join in

When your boss is hyping it up and demanding all hands on deck full steam ahead on the Good Ship AI, lots of people join in out of fear, particularly in the currently awful job market which is partly being ruined by AI itself.

Some of us just stay quiet, keep our heads down, and plug away using tools actually fit for purpose, like LSPs and refactoring tools.

Very few have the courage to stand up in a professional setting and say the emperor has no clothes.

brookst7• 10 months ago

This is true, but it is also true that traditionally, people whose jobs are most likely to be affected by technological change are not the most dispassionate, objective judges of the technology.

geon• 10 months ago

No developer job has yet been lost to AI. We are in a huge recession, and companies like to have an excuse to fire people that can be spun as a positive.

steveBK123• 10 months ago

Absolutely, and you see this in non-Mag7/non-tech industries where CEOs are all announcing a parade of "head of AI" hires. Will it end up being a permanent role like CISO, or is this just faddish follow-along? I know some of the guys being hired into these roles and, lol.. let's just say they are not experts.

My spouse works at a large (50k-100k) org in a program management role where she is getting a lot of pressure to organize various AI evangelism efforts aimed at developers. Workshops, bake offs, demo days, etc.

I mean sounds neat, but is this being done because it's useful or because someone up high needs to justify their AI budget spend with AI usage metrics?

Do we believe that ICs are actually so stupid/stubborn they need to be mandated, coaxed, coached, bullied and bribed to use something that makes their jobs easier?

Doesn't most of the best tech end up being bottoms-up?

Most of us who were around 15+ years ago recall a lot of BigCorp had to be dragged unwillingly into mobile by internal useres/devs who got their first iPhone and saw the light. A lot of stuff starts as small team internal skunk works / unofficial projects working around productivity drains. I am highly suspicious that the C-suite knows what people 10 levels down actually need for productivity enhancement.

darepublic• 10 months ago

I sympathize with this, I would hate having some cargo cult bullshit foisted upon me by disingenuous management. However I'm not sure that it's AI who is "the emperor with no clothes". What I think _is_ such a facade are oftentimes the company itself, its business model, the product it is trying to build and the way its trying to build it. I have sometimes spoken my mind about these things -- petulantly, on my way out the door.

geoduck14• 10 months ago

It reminds me of when people first got their phones and couldn't stop showing everyone how cool they were.

Your comparison to smart phones is interesting. Smart phones are definitely transformative. There was a lot of hype, but still transformative.

Do you believe that LLMs and AI is also going to be transformative?

brookst7• 10 months ago

People keep rediscovering the trough of disillusionment and mistaking it for a dead end.

pixelesque• 10 months ago

Likewise with the dot com bubble and the web - it was a bubble and it was overhyped, but it was still transformative if you look back 20 years as to how things are different in terms of media, and commerce.

redwood3• 10 months ago

I think the transformation will primarily be in search personally. As in Google search type experiences.

What that means is the ad model of the internet will come apart.

And what that means is that the LLMs will need to charge for answer optimization to plug the ads hole.

And so where this is going is basically a whole cottage industry around that. Around controlling and shaping knowledge in other words.

Yes frightening politically more so than economically. At least from my view.

And if it dumbs us down and erodes critical thinking then maybe it will have negative effects economically and politically long term.

pjc50• 10 months ago

Transformative, but not necessarily in a good way: likely to lead to the end of the open internet, along with all sorts of weird social effects from lowering the cost of convincing fakes.

tokai• 10 months ago

I don't think it was about smart phones.

zippyman553• 10 months ago

There are no error bars, no confidence intervals. Just a one trick pony that pastes tokens together to give you something that may look good to many people. Sure there are many good use cases, but there are still enough non-patchable and indeterminable in size holes in the watering pail to limit its effectiveness.

brookst7• 10 months ago

Are you saying the Internet is a fad?

simianwords7• 10 months ago

Is there any error bar or confidence intervals in stackoverflow answers?

qsort• 10 months ago

how in anonymous forums there's a lot more people pointing out that they think this is hype whereas when we wear our professional hats many of us join in

Different speeches for different audiences. On HN, for all its faults, people don't need to be told that yes, SOTA LLM can somewhat help you with code, parsing documentation, etc. A lot of people in the "real world" are still grossly underestimating this technology.

wffurr3• 10 months ago

> A lot of people in the "real world" are still grossly underestimating this technology.

Did you mean "overestimating"? "somewhat help" is putting it strongly, IMO.

baggachipz• 10 months ago

It's definitely interesting how in anonymous forums there's a lot more people pointing out that they think this is hype whereas when we wear our professional hats many of us join in. It's like we all want you to party going no we all know what's going to happen

Well, when people's financial and employment stability are dependent on placating the overlords who are entranced...

numbers_guy• 10 months ago

This technology demos incredibly well

You just summed up machine learning, not just AI/LLMs. My domain is very far from LLMs, but even in my domain, you can build a really cool demo, that is entirely misleading.

Fricken3• 10 months ago

I've been following progress on self-driving cars since the 2004 and 2005 Darpa Challenges. Things were very exciting in the 2011-2015 timeframe. Demos were mindblowing. Engineers were giddy. The auto industry went from laughing to shaking in their boots, and then scrambling to make it appear as though they had a horse in the self-driving car race. Hundreds of billions in investment were flying around. Now here we are in 2025, things aren't dead at all, but also nobody with skin in the game is under any delusions about how hard it is to build and scale a viable robotaxi operation.

So far the Self driving car hype cycle has served as a useful reference for understanding the hype with LLMs. The main difference with LLMs is that there is 10 times as much money flying around.

2muchcoffeeman• 10 months ago

But I can actually use an LLM to do useful work for me and no one gets hurt.

I think there’s too much hype and money, but LLMs are useful right now.

boplicity5• 10 months ago

I think it's easy to forget how much low-hanging-fruit there still is, in terms of taking full advantage of this technology.

People are still figuring out very basic integrations, and even now, at this early stage, the things I can do with LLMs are pretty incredible. For example, I was able to set Cursor up on finally dragging an old codebase out of the dark ages. It then built new features that I've long wanted. It took a few hours on my end, but would have been at least a week or two without it.

I'm not exposed to much of the hype, though, so maybe my calibration of what the hype is, is wrong.

chain0304• 10 months ago

"I think it's easy to forget how much low-hanging-fruit there still "

Such as?

noosphr• 10 months ago

On the one hand I see posts here by people who have no idea what they are talking about.

On the other I get 5 hours of work done in 5 minutes every other day.

Worst I can see happening is a doc-com crash. pets.com will go out of business, but amazon won't.

rsynnott2• 10 months ago

One thing I find fascinating; go to any forum/subreddit/whatever for any LLM thing, and it will be full of people complaining that it's not as good as it used to be, and that OpenAI/Anthropic/Google/whoever is intentionally degrading it, because they are so evil and want their products to be worse.

Then a new model or tool comes out, all is wonderful for a bit, then repeat (except for GPT-5, oddly; that one seems to have inspired hatred from the start).

It rarely seems to occur to people that familiarity breeds contempt; once the novelty wears off people start noticing the problems. The model isn't getting magically worse, you're just _noticing_ more.

boplicity5• 10 months ago

When GPT 5 came out, I was using GPT 4 mini for a regular automated task. This had been working quite well for some time. It stopped working right when GPT 5 came out. I switched to GPT 5 mini, and it started working exactly the same as it did previously. So yeah, I'm pretty sure they nuked GPT 4 when they launched GPT 5.

mvdtnz• 10 months ago

I can't speak for all models but I can tell you for an absolute fact that Claude has degraded recently. Yes I've noticed it subjectively but Anthropic have also straight up come out and said "lol oops we 'accidentally' made it way worse, now that you've all noticed we'll roll that back".

https://www.reddit.com/r/ClaudeAI/comments/1nc4mem/update_on...

thisisit• 10 months ago

It's definitely interesting how in anonymous forums there's a lot more people pointing out that they think this is hype whereas when we wear our professional hats many of us join in

I have seen that in anonymous forums there's a lot more people pointing out that they think this is transformative like say early smartphones.

When it comes to wearing professional hats people fall into two categories.

People who are using it outside their expertise area. Things they had to previously rely on others like back end operations. They gush about it - praising the technology as if they have discovered smartphones. Only when they use it for their expertise area they realize that AI might not be all it is hyped to be and that they need to be cautiously optimistic.

Then there are people who are using it within their expertise area. Like vibe coding. There are people who gush about it. But there are more who are cautiously optimistic. They will tell you it works but in a contained environment.

TNDnow• 10 months ago

Fiat money in a credit based inflationary environment is free. Oil money is free, it literally comes from the ground. There's no "losses".

wffurr3• 10 months ago

Not sure what this has to do with the article.

Oil isn't "free" anyway, it takes energy to make energy, and EROEI has been going down for some time as the easy oil is extracted.

lherron• 10 months ago

This “article” is clickbait. Controversial title with no substance, asking “why are companies investing heavily in a technology that works for some (limited but valuable) use cases, when they could invest in pure R&D for something that might be better someday”.

Mizza2• 10 months ago

What's the path to recouping that money?

Even if every major company in the US spends $100,000 a year on subscriptions and every household spends $20/month, it still doesn't seem like enough return on investment when you factor in inference costs and all the other overhead.

New medical discoveries, maybe? I saw OpenAI's announcement about gpt-bio and iPSCs which was pretty amazing, but there's a very long gap between that and commercialization.

I'm just wondering what the plan is.

HWR_14• 10 months ago

If LLMs could double the efficiency of white collar workers, major companies would be asked for far more than $100,000 a year. If could cut their expensive workforce in half and then paid even 25% of their savings it could easily generate enough revenue to make that valuation look cheap.

mathw2• 10 months ago

Unfortunately for the LLM vendors, that's not what we're seeing. I guess that used to be the plan, and now they're just scrambling around for whatever they can manage before it all falls apart.

Gigachad3• 10 months ago

Ok but then another AI company would just offer the same thing at a lower cost.

frfl2• 10 months ago

Wasn't the plan AGI, not ROI on offering services based on current gen AI models. AGI was the winner takes all holy grail, so all this money was just buying lottery tickets in hopes of striking AGI first. At least that how I remember it, but AGI dreams may have been hampered by lack of exponential improvement in last year.

brookst7• 10 months ago

I’m sure somebody believed that? But I never met them.

mathw2• 10 months ago

I think it's "scam everyone into giving us lots of money, then run before the bills come".

brookst7• 10 months ago

$100k/year is literally nothing.

Think of it as maybe $10k/employee, figuring a conservative 10% boost in productivity against a lowball $100k/year fully burdened salary+benefits. For a company with 10,000 employees that’s $100m/year.

techpineapple2• 10 months ago

But we won’t get there unless the company integration failure rate falls below 95%

Mizza2• 10 months ago

Even at $10k/yr/employee, you'd need 30 million people on the 10k/yr plan to hit 300B ARR. I think that's a hell of a big swing. 3 million, recoup over ten years? Maybe, but I still don't think so. And then competition between 4 or 5 vendors, larger customers figuring out it's cheaper to train their own models for one thing that gives them 90% of the productivity gains, etc.

But rather than speculating, I'm generally curious what the companies are saying to their investors about the matter.

34679• 10 months ago

That's literally not how the word "literally" works.

CyanLite2• 10 months ago

Major companies will spend 10-100x that if it resulted in real tangible productivity gains for their businesses.

ACCount37• 10 months ago

The "game plan" is, and always was, to target human labor. Some human labor is straight up replaceable by AI already, other jobs get major productivity boosts. The economic value of that is immense.

We're not even at AGI, and AI-driven automation is already rampaging through the pool of "the cheapest and the most replaceable" human labor. Things that were previously outsourced to Indian call centers are now increasingly outsourced to the datacenters instead.

Most major AI companies also believe that they can indeed hit AGI if they sustain the compute and the R&D spending.

elzbardico• 10 months ago

If AI Doesn’t Fire You, It Can’t Pay For Itself https://esborogardius.substack.com/p/if-ai-doesnt-fire-you-i...

simianwords7• 10 months ago

They just have to be positive revenue on inference and run it for a long time. Why do you think they can’t recoup it?

weego• 10 months ago

I think what we have right now is an excellent interface for low-friction, non-specialised interaction by humans (work or personal use) with a vast array of highly specialised and highly-complex systems.

What it isn't is the actual final "thing" itself. It's just the thin veneer right now.

I'm not convinced that that revolution was worth whatever trillions we'll end up spending, but fortunately that's not on my shoulders to be worried about.

shoo3• 10 months ago

How big is $344B?

Apparently the total market capitalisation of the US stock market is $62.8 trillion. Shiller's CAPE ratio for the S & P index is currently about 38 -- CAPE is defined as current price / (earnings, averaged over the trailing 10 years)

That suggests that over the last 10 years, the average earnings of the US stock market is about $1.7 trillion annually.

So $344B of spending is about 1/5 of the average earnings of the total US stock market.

Still hard to interpret that, but 1/5 is an easier number to think about.

brookst7• 10 months ago

But the $344B figure is not annualized, it’s cumulative.

shoo3• 10 months ago

the two nested bloomberg articles say

This year the world’s four largest tech firms will spend $344 billion on AI
Altogether, the four companies are expected to spend more than $344 billion for the year, with much of it going to the data centers necessarily to run AI models.

so both articles frame that $344B as estimates of capex within 1 year.

ghm2199• 10 months ago

These days am looking at managing my own portfolio. I go off expected returns(using gdp as a component) plus dividend and adjust it for risk to compute which country/region has good expected returns. This i adjust a couple of times a year.

If one would assume it's nearly all a bubble, How would you correct earnings for the US? I am interested in applying it to any investment that tracks AI heavy companies in the US.

shoo3• 10 months ago

If you believe in long term mean reversion of CAPE ratio for US stocks, you'd expect price/earning multiples to contract by a factor of 2, over some hard to predict time frame, where CAPE reduces from 38 back to about 20. If we arbitrarily guess that contraction happens over 10 years, that'd be -6.7% / yr for 10 years, from 0.5^(1/10). Then add the return components you mentioned from dividends and earnings growth.

One approach I've seen a few folks do is to fit a regression model of annualized real stock market returns over the next 10 years as some function of CAPE or 1/CAPE or log(CAPE).

It doesn't give a very good fit on training data, R^2 in the range of 0.2-0.3, i.e. it cannot "explain" most of the variation in 10 year returns.

CAPE based regression models like that have said the US stock market has been overpriced for the last decade! But investors in the US stock market have done pretty well over that period, with really good returns. Maybe these models are accurate but we've just gotten lucky? Maybe these models aren't very good. Hard to tell.

Elm capital publish estimates of expected returns of a few asset classes quarterly: https://elmwealth.com/capital-market-assumptions/

lotsofpulp• 10 months ago

How do you plan to adjust for government actions that affect purchasing power of the currency?

thiago_fm• 10 months ago

They MUST. It doesn't matter if it looks fragile or how much money it is.

LLMs risk most of those companies business, they can't afford to not be ahead. If they aren't ahead, there's a risk that the entire US's economy would be in a terrible shape.

American Big Tech companies that make plenty of INTERNATIONAL revenue from Ads (Meta, Google), can quickly become a shell of its former self.

How? Countries and economic blocks could quickly substitute their American products counterparts if they have nothing to offer and could roll out their own.

The US's economy has become very dependant on FAANG cashflow, it's what gets other parts of the economy moving.

No wonder they had a dinner with Trump. If this fades away, US will look very weak and with a terrible economic outlook.

kingkawn• 10 months ago

In the same way people thought the train would hit them when cinema first debuted so too do we believe the machine is thinking

gsky2• 10 months ago

Most of us believed that crypto currencies were trash. Look how valuable its now .

LLMs are a million times better than Crypto currencies.

Gigachad3• 10 months ago

Crypto never ended up filing those original goals of being an actual currency and smart contracts actually doing things. Blockchains and NFTs were never used to solve any of the problems they claimed to.

It’s just none of that had any baring on the value of the coins.

bjacobel• 10 months ago

There is a distinction between "vehicle for a large volume of speculation" and "valuable". Cryptocurrency has not created any "value".

lm284692• 10 months ago

Pokemon cards are also super valuable these days, it says more about some people having way too much money for their own good than anything

ninkendo• 10 months ago

"LLM's are the next crypto!" isn't the glowing endorsement you seem to think it is.

gsky2• 10 months ago

Such a world we are in. Nothing can be predicted with certainty.

LLMs will be a million times more valuable than crypto shit in my opinion.

impossiblefork• 10 months ago

It's obviously a lot of money, and of course it's too much money, but I think we can still get LLMs much further and I think they're probably the currently most interesting approach.

I don't even care about multimodality etc. I think pure text models are a very appealing idea.

rsynnott2• 10 months ago

Hallucinations haven’t gone away, muddying the path to adoption for companies in healthcare or legal analysis

... I mean, of course they haven't. They are a natural consequence of how the things work!

pmdr4• 10 months ago

One can call into question the paucity of AI-critical posts & comments here on HN. Not much is being said about the economics, everybody's living off the hope that "they'll figure it out." And AGI is just a few hundred-billion-dollar loans away, so why quit while we're ahead?

If they do "figure it out" (both AGI and a viable business model), a lot of people here will likely be out of a job. If they don't, the whole thing will come crashing down, taking our invested savings with it.

The comparison to crypto keeps coming up. Not everyone's savings went into crypto, but a lot of people's savings and retirement funds are being invested in funds tied to the stock market. And right now its growth depending on pumping cash into the AI bubble.

simianwords7• 10 months ago

The answer is so easy to solve.

Do I find value in paying 20 dollars a month for ChatGPT? Yes. Do others? As far as I can see, yes. Most people are happy with the value.

Are AI companies profitable if they stop R&D? Yes.

Where’s the skepticism coming from?

pmdr4• 10 months ago

I'd actually be content knowing $20 a month covers inference and makes them a decent profit, but I have my doubts. If there's proof out there that it does, I haven't found it yet.

Are AI companies profitable if they stop R&D? Yes.

By what metrics? Where's the data?

ankit219• 10 months ago

there is clearly overhype. Given the influx of people building here (and i am not aware if it happened previously too), in a bid to differentiate from other startups building the same thing, many just stripped the nuance out of any technical idea, and made it a simple marketing term. As an exec where ten startups are promising you "training on your data" to provide the best chatbot, it's hard to tell the difference between who would be actually training and who would be just tweaking prompts. This has happened to other concepts too (anecdotally, most famous is how everyone is offering deep research). yes, it helps growth, but comes at the cost of trust. There is this startup which promised "experience based learning" when all they were doing is adding memory to the prompt to get it to perform better. (you can look it up, recenty raised series A).

This does not mean ideas are not working. I personally think pretraining has done its job. We did not know what the job previously was, but now we do given the way RL works. Pretraining and test time compute enables models to develop a generalized prior they can use to solve any given problem (much like how humans solve such problems). Sometimes priors are lacking so you need to train more using RLVR, and still early days, but directionally I think we have another scaling curve here.

AI's $344B 'language model' bet looks fragile

Related Stories