A while ago I was thinking, “Gee, AI is so complicated, how can I keep up with the landscape?”
After reading these articles go by so often, it feels like what I actually can’t keep up with is the bond market. To paraphrase Trotsky, you may not be interested in the bond market, but the bond market is interested in you. I want to be able to read the signals at the bottom of this article, and divine some kind of prediction that can guide me… I don’t know, to choose whether I should buy a house or change the investment strategy in my retirement fund or something. But I’m just seeing all these signals go by, waiting for the story to be written, which only happens when the dust settles.
I guess I’ll go back to not understanding AI, instead of not understanding the bond market.
As a retail investor, you should buy an index fund and then forget about it.
By the time you have read this article, the professionals and their computers will have digested that material a thousand times over and have priced it in.
To be more precise: buy the lowest cost most diversified index fund you can buy and then hold it. If you want to spend some smarts to get a better return: look at how to minimise taxes and fees.
> By the time you have read this article, the professionals and their computers will have digested that material a thousand times over and have priced it in.
I think you overestimate traders. What we call smart money is very often really, really dumb from a macro perspective. Professional traders believe hype and follow trends. There is still at least 2 thesis playing out at the moment for the AI trade, and you don’t need to be a professional trader to take part: one is the AI impact on saas (the market has been very bearish on SaaS companies, and still hasn’t corrected meaningfully), and the ai infrastructure (hardware companies + hyperscalers)
The average professional trader might not be that smart, who knows. But:
(A) I am not confident I am smarter than the average professional trader here.
(B) You don't need to win against the average professional trader: you need to win against the smartest ones. And: I'm not so sure I am smarter than the average professional trader anyway.
About your theses: my null hypothesis is that these things are already priced in.
You actually don't need to win against the smartest ones, it's not really who is on the other side of the trade. There is an interesting dynamic where the space where retail investors operate isn't really the same space as the "smart money", and because of that there are a lot of small-ish arbitrage opportunities that you can advantage based on your industry expertise (I don't mean insider trading, I mean your understanding regarding how your industry is evolving). Relying on an ETF is perfectly fine, it's way less stressful and gives you the safe-ish exposure to the market most people want. But if you want to do active trading by picking stocks (and you don't mind risking losing your money), you really don't need anything too sophisticated.
I like it how morningstar provides a graph of the stock price and morningstar's target price for the past few years, and the target price always closely follows the stock price, even as the stock price fluctuates wildly. All that genius analysis somehow concludes that the company is worth exactly what the market says it is, right up to the point that it doubles/halves in price.
For the record, I deliberately picked an index fund (VWRA) that's not choosy about who to admit, so SpaceX would have been in there pretty quickly no matter what shenanigans they are doing with the S&P500.
> As a retail investor, you should buy an index fund and then forget about it.
That definitely was true.
I am unsure it is still true. Index funds have taken so much of the trade volume the are becoming momentum strategies
So long as you are happy following the market wherever it goes, and if the recent past is a guide then up is the direction, then yes.
But given the nepotism and corruption in the highest reaches of USAnian society (e.g. Trump's crypto currency scams and the blatant inside dealing and rule ignoring of the Space X float) the future looks much less certain than the past
> I am unsure it is still true. Index funds have taken so much of the trade volume the are becoming momentum strategies
I don't understand how that's supposed to work?
Btw, keep in mind that index funds are typically really, really keen to lend their shares out to short sellers.
> But given the nepotism and corruption in the highest reaches of USAnian society (e.g. Trump's crypto currency scams and the blatant inside dealing and rule ignoring of the Space X float) the future looks much less certain than the past
That's a big part of why I am invested in a global index fund, not anything America specific.
Depending on jurisdiction and taxation, yes. Eg in many places owner-occupied housing is favoured, and might make sense to acquire, even though otherwise it's silly: a single lumpy usually highly levered position; no diversification; multiple times your networth and with very high transaction costs.
I think there is now so much passive investing that wall street and tech bros are gaming it, and it's no longer based in reality (fundamentals). I mean Elon's deal to get SPCX short listed with NASDAQ was directly targeted at 401(k)s.
Not sure how it's all going to play out, but this ginormous increase in passive investing over the past decade or so, mainly in S&P 500, seems like a vulnerability. Small cap might be a better (non-sexy) target long term.
> I think there is now so much passive investing that wall street and tech bros are gaming it,
Yes, they are supposed to! And they are supposed to compete for the privilege. That's how index funds can add and remove stuff from the index so cheaply.
I agree that the S&P500 is not an optimal index. I picked something (VWRA) that's more diversified and less picky about who to admit.
Don’t forget there is a constant pressure for everyone to have an opinion on how this is going to end while hoping it ends tomorrow so they can be vindicated. The dotcom bust took a decade to grow and collapse. I think it is too early to make predictions with AI. I mean the sentiment here is either it will dry up the world and kill us all or transcend humanity, there’s no gray area. I don’t want to fall into the emotional sieve that seems to drive everything.
That's the financial stakes here. That's why it's all or nothing. You're spending on a level that is only justified by the bonafide machine god being ushered into existence, not productivity or coding tools (and on relatively short time horizon). So if this doesn't change the near term trajectory of humanity to a parabolic move upward there is going to be a lot of economic pain. It's not just the spending, it's that the expectations for the returns to justify them are in a relatively short period of time.
> if this doesn't change the near term trajectory of humanity to a parabolic move upward there is going to be a lot of economic pain
"trajectory of humanity to a parabolic move upward" is poorly defined here. Whether we are headed to a machine god ruled scenario or "just" incredibly powerful productivity tools, there will be a lot of economic pain for some (most) and a lot of economic gain for a few.
I've yet to a see an LLM/agent-based business plan in where scaling with an order fewer workers than before LLMs is not a central part of the value proposition.
such a business plan has not yet created economic value, being able to roll out features at rocket speed is not a huge determinant of a startup's success.
> such a business plan has not yet created economic value, being able to roll out features at rocket speed is not a huge determinant of a startup's success.
Speed of feature implementation isn't the objective. High feature throughput with labor cost reduction is.
I think these large numbers are casually thrown about, but the real meaning is mind boggling. 1 trillion dollars is the entire US defense budget - aircraft carriers, nuclear submarines, health care, salaries, stealth fighters ect. The hidden AI debt alone is more than that https://asia.nikkei.com/business/technology/five-us-tech-gia... just for five tech giants (not to mention all the other smaller players like neoclouds)
Most people mean this to say that 1 trillion is a lot of money, but it still comes back to what you believe AI is- in hindsight, does 1 trillion dollars to build the internet sound like a lot or a little? (That is, spending 1 year of USA's defense budget to get the entire internet)
It comes back to your perception of what AI is because to people who say AI is glorified auto-complete won't believe that the money is worth it.
The AGI-pilled true believers who say it will end all money and result in a post-scarcity world believe literally any amount is justifiable.
Most people, me included, land somewhere in the middle- it seems like AI is a humanity-level sea change in technology and how computers work and serve us. It seems plausible that a few trillion is a reasonable amount.
I'm not going to make a prediction of what will happen with AI whether it will autocomplete / productivity or AGI. I will say it seems to be trending towards former than the latter just by how scaled down the promises have become over the last year (we went from curing all disease and cancer / post-scarcity to productivity and code.) The amounts being spent on this can only really justified by some paradigm shifting returns and within the timeframe investors expect. This isn't something like Apollo / Manhattan project - those were taken on by the government with public money. This is explicitly a profit making enterprise funded by markets.
1T is big in the absolute sense, but that's simply the scale these big tech companies operate at. Go back to 2024 or 2025 and you'll see as a group they are making a net income of $400B+. The scale at which these companies do anything is just staggering.
Good that you made the comparison with defense budget because it's becoming evident that AI is the new "nuclear bomb". First, if you have more advanced AI than your opponent, you just hack them and the war is over before it began. I'm simplifying obviously, but the point that AI is essential to modern warfare stands. Second, even during peacetime, you can use AI to directly influence what people think, what they do, etc. in the most literal sense of this expression. We already know that people outsource thinking, relationships, and cultural expression to LLMs, and even if that wasn't the case, you can use AI to deliver to each single person hand-crafted propaganda, not to mention the previously unimaginable opportunities to spy on people.
It would be strange if the race to wield this power wouldn't result in AI getting pumped to the moon, way beyond anything that seems reasonable.
That could all be true, but the problem is however powerful it is, it still needs to make money for the people who invested in it who are expecting a return. The stock valuations, the bonds yields - they don't care about the power of a nuclear bomb. They want to get paid. They expect to get paid. And if they don't get what they are expecting, there will be hell to pay in the economy. There's a probably a good reason the power of nuclear weapons isn't an ETF I could buy into.
I mean that's so far, it continues to grow exponentially larger with each quarter. The debt issuance for the first half looks to be crowding out US treasuries in the bond market - https://www.bloomberg.com/news/newsletters/2026-07-23/ai-deb... - that's an extremely large amount of debt. And it's still getting larger and larger each quarter.
Yeah this is going to be north of 10 trillion by the end, I would wild-ass-guess. Inflation adjusted it's larger than the manhattan project, apollo program, works progress administration, hell, it's on par with the cold war era military buildout, or a baby world war.
I would question the idea that highly industry consolidated debt is competing with risk free debt issued by the US government. Those are two very different products.
And while the quarter by quarter growth may seem astonishing it very different saying “debt levels today are alarming” versus “if this trend continues debt levels will be alarming”
So the disclosed balance sheet debt is 1.35 trillion and then the off-balance sheet debt is 1.65 trillion for a total of 3 trillion in AI debt for the 5 tech giants so far. It's multiplying every quarter and they've set investors expectations to be that this is never ending basically. But the tech giants aren't the only people spending themselves into massive debt, think of the CoreWeaves and the Nebius and the hundreds of other smaller companies. And the expectation is that there will be a near term return on all this with a healthy profit. Those five tech giants are just the tip of the iceberg in terms of the amount of debt.
While the debt numbers might look large, don't forget how much revenue is coming into these companies as well.
OpenAI gross revenue was $4B in 2024, $13B in 2025 and estimated to hit $26B in 2026. Revenue is doubling or more.
Same with Anthropic. $1B in 2024, $9B in 2025 and estimated to be $47B in 2026.
All the AI companies together are pulling in hundreds of billions of dollars per year in revenue and it's going up quickly. To me, taking on trillions in debt in order to make hundreds of billions in revenue doesn't seem crazy.
What’s weird is how emotional people get on this. I told publicly (because I was asked, not out of an obligation to have an opinion), that the prices we pay for LLMs are likely to go up because that’s what happens when the ratio of operational assets to foreign capital drops due to the capital having been turned into heat rather than operational assets. The grief I got from people, dear Lord…
I think that opinion is as reasonable as any. I feel compelled to argue against it (I even thought out the arguments in my head!) but my compulsion to have an opinion on HN is a disease, and you made a point of saying that you gave the opinion because asked.
I feel obliged to step in here to say there is a grey area where these are useful tools for some applications but not on the path to AGI.
Unfortunately the hype machine has far outstripped their capabilities so far, and the amount of money spent doesn’t look like being recouped, so somebody is going to lose money, as people lost money on the overpriced spacex ipo (overpriced because of AI).
> The dotcom bust took a decade to grow and collapse
I'm inclined to think the collapse has already started but nobody wants to see it yet.
In the last few weeks SP500 is down, kospi is down, nikkei is down, US inflation is still high and growth is softer than expected. Hyper inflated stocks (Tesla, Nvidia, SpaceX) are deflating. US bonds are at a 20 year high.
> Gee, AI is so complicated, how can I keep up with the landscape?
The interesting thing is: you do not need to keep up. It’s actually way easier and cheaper to wait a bit for the chaos to stabilize, then learn to use the tools. You don’t need to have been someone who experienced the whole evolution, non stop at the edge. It’s ok to let the enthusiasts discover how things work and eventually learn from them. Just like any other technology. The whole „you will be left behind“ is nonsense. If AI is the future, then it will here to stay and you can let others map the domain first
I think there's a common misconception at play here: using proprietary software thought of as a skill
Must be a mix of decades of corporate branding in software and the erosion of trust in the labor market coupled with the rise in noise from various sources...
Skills are what they always have been. Logical thinking, good memory, long attention span, critical thinking, communication (the ancient Mediterranean, the culture which forged the currently dominant "West", valued 'rhetoric' almost above everything else besides physical prowess)
Being well connected, having a sharp mind, cultivating deep knowledge of science, philosophy, language, mathematics. These things are timeless. Using the current brand of software well is a moot point. Software can only become dominant if it falls in line with mainstream design. Learn design, you can intuit how to use relevant software and the badly design tools will be weeded out naturally.
In a nutshell, exercise body and mind. Be strong, cultivate strong relationships. Everything else is just background noise.
I was able to create a custom index based on the top 500 that I stripped the big AI stocks from (shovels too). Then I added decent chunks of international, small cap, and treasury ETFs to it.
I have no illusions that I can time a bubble, but I'm hopeful I'm at least partially shielded, and most importantly I feel better about ignoring wall street again.
Instead of starting with the top 500 American stocks, and then adding international and small caps, you can start with a global all-market stock index--and then remove AI from that.
Yeah I couldn't figure that out with Questrade (Canada). It's a pretty new feature, but I think it's great, so I hope they expand their baseline indexes.
I was considering writing a tool that simply follows any index you choose with a .toml of simple config options, like which stocks to exclude, potential fixed locks for specific stocks (or maybe upper and lower percentage of portfolio settings), a hard per stock cap (say AAPL at 3%), and drift threshold. Something you just run once a day and it spits out your buy / sell orders. Seems like this is something brokerages are already offering in some variation though, and I'm not sure what, if any, API access looks like, or export / import options.
I'm in Singapore. My money is in VWRA (without bothering to remove AI companies).
Your idea for the tool sounds interesting. I suspect even just copy-and-pasting the paragraph you wrote here into your favourite AI programming agent would get you pretty close to a prototype you can play around with. At least in terms of 'spit out buy / sell orders' and leaving out the API integration.
Yes, it's a very simple concept IMO. Without API access or at least CSV import / export integration w/ a brokerage for automation I don't think I'd use it. I could have an agent use the Web UI on my behalf, but honestly, that feels like lighting tokens / gas on fire.
~ 5% Nvidia as biggest holding and 20% in US listed tech companies (most of which are heavily invested in AI), over 60% in the US market, so this ticker is very similar to investing in the US market alone.
Also when a bubble like this deflates it hits almost everything so it is very hard to avoid, but world indexes are particularly exposed.
You are right about the numbers. But wrong about what I am thinking: I'm aware, and it's a deliberate choice on my part to allocate in proportion to market cap.
Revolutionary technology + massive adoption ≠ good investment
Investors have poured money into a bottomless pit, attracted by the growth and glamour of the industry. The airline industry since its birth has had a collective net loss, in aggregate, despite moving hundreds of millions of people.
Commodity Product, no switching costs. Infinite competition
The airline industry since its birth has had a collective net loss, in aggregate, despite moving hundreds of millions of people.
The industrialisation essentially socializes the cost across a lot more people though, so even though it doesn't make a profit it does mean people can have air travel without it costing millions per flight for the few people who can afford it. Essentially the economies of scale from having lots of flights isn't enough to make it profitable but they are enough to make it affordable.
There's no spare money to extract from the airline industry but it's still very useful. The same could be true for AI in the long term.
Sometimes the goal of an industry is to exist rather than to make a profit, because the benefit to society is more important than profit. People don't like that though so they do a bit of creative accounting or head-in-the-sand denial around it.
> There's no spare money to extract from the airline industry but it's still very useful. The same could be true for AI in the long term.
Of course it could, let’s start with making the models open weight and entirely open source. Fully publicly owned and not shaped to maximise profits for the shareholders.
Oh wait, Scam Altman entered the chat and turned a non-profit lab into the next biggest IPO vehicle the world has ever seen.
OpenAI launched as a nonprofit research institution. Its announcement explicitly said it wanted to pursue AI “unconstrained by a need to generate financial return,” produce value for everyone rather than shareholders, publish research and share patents broadly.
in the case of airplanes the only thing thats the private market is the planes and the ticket, the entire system of airports, safety, navigation is state subsidized and when the market fails it gets bailed out. the oil is subsidized by constant warfare. it's just an illusion for reganomics so a few rich ppl can make a buck off of a public utility.
How does constant warfare subsidise oil? In case you haven't noticed: both the latest US-vs-Iran war and Russia-vs-Ukraine war have made oil and gas a lot more expensive than the peaceful counterfactual.
It isn't a commodity product in my opinion. Far from it. I think it will ultimately be a monopoly or duopoly for SOTA. The mid to low end is commodity, yes. But SOTA models are not commodities.
The number of competitors for SOTA drops by a few every year. The winners make more money, get more revenue, buy more compute, train better model with compute, buy best talent, and the cycle goes.
I think it's easier to fall behind and never catch back up than people think. One disastrous training run can leave a lab months to a year behind. For example, Meta's disastrous LLAMA 4 models. Meta is lucky to have their ads business as a funding source. However, Anthropic's revenue is growing so fast, that ability to use ads as a funding source to stay in the race may not last much longer for Meta.
To me, SOTA LLM training is very much like new chip fab nodes. One disastrous node can put you behind for many years or forever. The cost to build the next chip node doubles every every 4 years (Rock's law). The cost to train the next SOTA model likely has some similar power law which means over time, it's too costly for losers to keep up. The only reason TSMC isn't a defacto monopoly for advanced chip nodes is strictly due to geopolitics.
All things equal, let's say your SaaS startup uses GPT 5.0 (release 10 months ago) and my business uses Fable 5. We have the same business goals, same talent level, same strategies. I think the chance of my business winning against yours is higher.
It's easy to agree to that, but you're disregarding that the resources you spend on the stronger model could be allocated elsewhere. Conversely, you're assuming that spending more on AI will always yield better results and be worth it, compared to spending the money on other things.
This might actually still hold true now, or or at least many actors in the market behave that way. But I'm not so sure there isn't a cliff to that effect. At some point, if SOTA models remain expensive, it'll turn into a market advantage to figure out how to get things done without depending on the most expensive tooling available.
Similar scenario, different phrasing: if your company relies on overqualified workers to deliver 100% quality, the market may still decide that it's fine to go with 90% quality for 50% the price.
@orwin has claimed that SOTA LLMs have already hit that diminishing return where spending more money on a SOTA LLM today does not add more value than a non-SOTA LLM (assuming high value tasks).
I never said there will never be a diminishing return. I'm challenging the statement that we've already hit.
Note: We're still scaling chip nodes. It's still worth it for TSMC and chip design companies to invest hundreds of billions into every new chip node every 2-3 years. This is after decades of scaling already.
I was thinking about Opus 4.5 when i talked about 10 month ago, but maybe i'm confusing the date, it might have been 9 month.
Opus 4.5 vs Fable 5? i think the output difference will be negligible, and the opportunity cost of paying an order of magnitude more will actually benefit my business.
I'd guess it depends on the type of business. If it's some genuinely deep technical space, the model would give an edge but even then I think luck would be a significant factor. In a monte carlo of such scenarios, the business with the stronger model might win 6 out of 10 times, but it's no sure thing between the two of us. If we were comparing two businesses building Yet Another Generic CRUD, I would guess it's closer... perhaps even a net-negative to spend money on Fable versus marketing.
Your business is spending way, way, way more for compute. How can you ignore the economic of your business? That’s pretty much what a business is about
That’s nonsense and saying “nah-ah” with Latin won’t improve your argument.
If we couldn’t isolate a variable we would never be able to argue.
Using a better model is an advantage even if only for the coders. There are a million ways to turn that into profit, both proper and not so proper but that’s the beauty of ceteris paribus: the other factors do not matter now.
> There are a million ways to turn that into profit
Mind pointing where that profit for companies consuming AI is? I don’t mean hypotheticals. Where are the proof that current AI contributes positively to ROI?
How can people in Hacker News still doubt AI's benefit when they are seeing in front of their eyes every technical profession getting disrupted to oblivion in the last year. Just ask basically any software engineer how much their profession has changed over the last 12 months
Obviously there is risk, but can't we really extrapolate the AI gains forward and just see how big it's ahead to become?
All I see is a flattening of the technical curve. Which is great, but the number of people who want to download an app is still the same. So all you have is 100,000 apps with no users instead of 10,000 apps with no users
You increased the amount of code written by 10x but unless there’s a 10x increase in demand, its worth nothing
I don't doubt that AI has benefits, but I do doubt that the major AI providers will be able to make back their investments. They've spent trillions of dollars, and yet they've barely created a moat. We're seeing open weight models being released that are only months behind them, that can run for way cheaper. This makes the future of OpenAI and Anthropic suddenly look rather bleak.
A simple analogy: If you have kids, you love them and want to give them whatever makes them happy. But on the other hand, you run a household, you pay for bills, healthcare, heating, education, and heavy overhead. You must keep things under control. You don't hand a blank check to an immature child who doesn't even know how to manage that money yet, right? So, even if your child wants to push forward at an extraordinary pace, you have to keep a level head, manage spending, and ensure everything doesn't end in ruin.
That’s the point: making growth sustainable over time.
Not seeing any significant disruption showing up in revenue etc from inside the industry, there’s growing skepticism about these tools and the claims made for them. They can be useful but are not world changing or actually replacing jobs.
We’ll see how they develop but so far they are not capable of operating independently.
I know many companies are spending quite a bit of money, I don’t know if it bears out that the increased spend has resulted in increased profits, even if there has been some increase in productivity. I think this is the tough situation many orgs are facing right now, drastic adoption without material economic gains.
How can people not trust in anecdotes and vibes while avoiding studies, do you mean?
Isn't that the point here? That everyone thinks massive disruption is happening and everyone is 100xing their productivity, but it's not actually showing up in the numbers anywhere?
> How can people in Hacker News still doubt AI's benefit
Because what many of us are seeing is meaningless “productivity” improvements.
If at the end of the day you don’t have more users paying for your product or the same users and paying more, then what’s the point of being more productive?
Lenders are doubting their return. People's benefits have nothing to do with it. The benefits would go in a minute, if doing so yielded a better return.
People are being a little unfair to you here, I think. I think there no chance of AI not being by far the biggest technological shift in our lifetimes. BUT that doesn’t mean any of the current companies leading the charge have sustainable business models, or that the current financing around it makes sense. Other commenters have pointed out both the railroads and dot-com boom as analogies, which holds up well. Generative AI is here to stay, but that in no way means that Anthropic and OpenAI are
I’m not sure that’s quite the right framing. If Anthropic goes bust, Fable persists as an asset that can be run by someone who didn’t have to pay to develop it, probably profitably, and probably in a way that gets cheaper over time. The debt pony show is paying for the next model.
> Grey Swans: risks that were in the data but overlooked or dismissed because few had synthesized the signals into a coherent picture.
Directly conflicts with
> Alert and Critical signals represent readings that have historically been associated with meaningful financial stress.
These are all pretty standard things to track and are regularly (and publicly!)
Not saying we’re not in a bubble or near/far from it popping, but these metrics aren’t going to precisely tell you _when_, which is pretty much the only thing that matters.
First, and obviously, the article is talking about capital when they say 'money'. Not all capital is borrowed.
Second, not all money is created via borrowing (but the vast majority is!)
And the YouTube video you linked to is very confused even about the money that is created via borrowing.
Government debt is not required to create money. The Bank of Japan bought stock ETFs to get 'freshly printed' money into circulation. ('Freshly printed' in scare quotes, because these days it's just entries in a database.) Another example: Singapore's central bank (MAS) does not use Singapore government debt to create Singapore dollars; I'm not even quite sure they would even be allowed to.
You can say that money itself is a debt of the central bank; and that's sort-of true, but it's not what David Graeber talks about.
A bit of a pedantic last point: silver coins or bitcoin also require no borrowing to create. Silver coins have been used as money, bitcoin could conceivably be used as money. (There are other problems with these options, but that's besides the narrow point.)
“None of this yet resembles a credit cycle turning, and the signals that would show one are quiet… What the data describes is closer to the opposite of a contraction: a credit expansion absorbing record supply and charging progressively more for it. The pricing runs in a ladder.
…
Where the ladder breaks is at the bottom. CCC and Lower Option Adjusted Spread signal rose twenty-two points over thirty days to 88 and sits at Critical, markedly higher than the Investment Grade or High Yield spread signal levels, and AI paper does not price at CCC. Private Credit Stress sits at 96, up thirty-one points, consistent with reporting data center financing has moved toward private credit and off-balance-sheet structures where the ultimate holder is harder to identify. What separates an expansion from a contraction is not the level of spreads but whether new issuance keeps clearing. An expansion growing more expensive still places its paper; a contraction is when deals stop pricing at any spread. The market is still clearing. It is clearing at a price that has moved in one direction.“
The real cost isn't the inference bill, it's the senior engineer hours spent debugging subtle hallucinations that slip past naive test suites. We stopped bleeding money on trivial boilerplate by implementing model routing through FlintAPI, ensuring only complex logic actually hits the expensive frontier endpoints. Treating LLMs as a monolithic tool instead of a tiered pipeline is what quietly destroys your margins.
you won't get debt if you don't have assets that can be repossessed, so having debt means these AI companies have assets: that's a strong thing, not a weak thing. interest rates are what they are, and they go up and down for reasons exogenous to your industry; debt regardless of interest is always "cheaper" than equity, and the shareholders expect to make their money from equity, paying interest on debt as a type of impedance matching and cost of keeping more equity.
so everything is going according to plan, and nobody knows the future, and predicting collpses has never been a profitable business.
I didn't have to read past the first few confusing contorted and convoluted paragraps of this article to decide to come over here and explain it, this is all straightforward corporate finance 102 and the article is fluff
I agree with the general sentiment, but I feel like it is also a bit reductive. Assets in this space are near impossible to evaluate and can fluctuate in value greatly based on other actors. In a hypothetical scenario where, say, google releases a new frontier model that somehow leapfrogs the competition by 5 months all of a sudden the value of the Asset of Fable 5 and GPT 5.6 might completely crater.
I'd argue that data in this case is more like the actual models they use, their codebase and their engineering talent. Not deep enough in the sauce to say one way or another how big the realistic delta between companies is though.
GPUs have a five year lifespan before they become obsolete and start experiencing reliability issues. We're already 1-2 years into that five year lifespan.
The payback time for a GPU running 24/7 inference is ridiculously short though. As little as 6 months according to some calculations. Most of that 5 year lifespan it will be earning well in excess of its replacement cost.
I would imagine Anthropic et al. are largely leasing land/buildings, so as the other commenter said… must be the server racks that are acting as collateral (if anything). Generally enterprise hardware depreciates very harshly. I’m used to paying $10 for Intel Xeons that once retailed for over $5,000. I expect to pick up some NVIDIA Blackwell 6000s for $100 each someday.
Yep, a friend recently told me that he remembers working somewhere that gave away old empty server racks - they were unnecessary, and expensive to store, so why keep them?
We are in odd times however - I for one am sitting on paper profits on the consumer gpu I bought 2 years ago. If anyone goes down before the supply side is fixed - the first to fall will probably be able to liquidate their gpus at a profit.
That wasn't what it seemed like at the time. Amazon didn't post profits, sure, but they sure as hell weren't a giant money suck either, they didn't need billions in financing to run their business. There were a lot of Amazon bears, but they were concerned about the high valuation, not about them going broke (since even the most pessimistic bear can read a cashflow statement).
And then everyone would stop using their inference as soon as a better model for a reasonable price came out.
The R&D expenditure is a critical requirement for the inference profits, to the point where we should probably lump their financials together, at which point is definitely not profitable.
What will it look like when R&D plateaus (and yes it definitely will, but it could take a while), investment falls, and a few main competitors remain in the music chairs?
It's very difficult to predict. The inference profits we are seeing the profits of a company that is temporarily ahead, but the revenue will level-out in a more stable market, depending on how many survived. It's also hard to tell where the costs will be at the end of the game, with constant efficiency optimisation mixed with cost increases for higher intelligence.
I think it will be quite similar to the semiconductor industry, where, yes there are some key monopolies, but they are not the initial big players, and none of it is actually very profitable; while the real profits are reaped by those that make popular consumer products based on the foundational tech. I guess the main difference is that OpenAI and specially Anthropic have been quite effective at directly tapping into the consumer market rather than remaining technology providers.
And then everyone would stop using their inference as soon as a better model for a reasonable price came out.
Exactly. It's competition now that is driving high training costs - not a business model problem. There will be winners and losers. The losers won't be able to keep up with the training costs forever. See my post here: https://news.ycombinator.com/item?id=49119265
It's never that simple, that's not the only possible endgame.
We have seen plenty of examples in other industries where you can never really stop investing a ton on R&D with diminishing returns (like in semiconductors or pharma), because the moment you stop newcomers overtake you.
Or the whole thing becomes a commodity with lots of competitors, where technological advantage is overtaken by marketing as the dominant force.
Or you really are left as the only player alive, but you realise that the market cannot absorb higher prices for your product by then, they prefer just not to buy it. Perhaps you are the only player alive because the business has become so low-margin that everyone else has abandoned it intentionally.
That Silicon Valley pitch you are echoing rarely works out as advertised, even for the winners.
We have seen plenty of examples in other industries where you can never really stop investing a ton on R&D with diminishing returns (like in semiconductors or pharma), because the moment you stop newcomers overtake you.
In semiconductors, it almost always become a monopoly or dupoly. x86 CPUs - only AMD and Intel left. Discrete gaming GPUs - only Nvidia and AMD left. 5G chips - only Qualcomm left in western market but Apple is about to join the part. In advanced chip node - only TSMC but Samsung and Intel survive due to geopoltics.
I think you're proving my point. Eventually, R&D heavy industries almost always become a monopoly or duopoly. Small/losing players can't keep up and drop out or acquired.
Check their R&D expenditures and profit margins they are quite poor.
Their revenue has always been sustained by the fact that their technology needs to be constantly replaced because it keeps getting better. The moment it stops getting better, the replacement rates plummet and so do their revenues. It's also really not that hard to compete with them when they get complacent.
I'm not counting Nvidia because they don't produce semiconductors themselves, they are a different kind of business.
They are indeed an example of those that cater to consumers and/or build popular products based on foundational tech from others, like Apple or Sony, which do tend to be quite profitable.
But the actual deep-tech semiconductor firms? They may be critical to the world economy, but they don't actually make that much money comparatively. In many cases they are not real monopolies, it's just that no one else wanted to continue investing in a shitty business model. Only the likes of TSMC and Samsung were okay in playing the low-margin game, but most US players left the board.
I believe AI has a lot of the same characteristics.
For how long though? If Amazon never built AWS the core business conceivably would still be around today, if Anthropic stopped providing new models two years ago no one would care about them now.
Amazon had a close call around the .com crash as capital markets froze, but they were not going broke every year. They were purposely (and rather famously in business circles) investing every dollar made in order to grow the business. It was clear early on the original business worked.
Amazon also added/pivoted to AWS, which is where a huge part of its value comes from today.
A while ago I was thinking, “Gee, AI is so complicated, how can I keep up with the landscape?”
After reading these articles go by so often, it feels like what I actually can’t keep up with is the bond market. To paraphrase Trotsky, you may not be interested in the bond market, but the bond market is interested in you. I want to be able to read the signals at the bottom of this article, and divine some kind of prediction that can guide me… I don’t know, to choose whether I should buy a house or change the investment strategy in my retirement fund or something. But I’m just seeing all these signals go by, waiting for the story to be written, which only happens when the dust settles.
I guess I’ll go back to not understanding AI, instead of not understanding the bond market.
As a retail investor, you should buy an index fund and then forget about it.
By the time you have read this article, the professionals and their computers will have digested that material a thousand times over and have priced it in.
To be more precise: buy the lowest cost most diversified index fund you can buy and then hold it. If you want to spend some smarts to get a better return: look at how to minimise taxes and fees.
How does 'the market' properly set prices if everyone blindly buying the same things without any discriminators?
> By the time you have read this article, the professionals and their computers will have digested that material a thousand times over and have priced it in.
I think you overestimate traders. What we call smart money is very often really, really dumb from a macro perspective. Professional traders believe hype and follow trends. There is still at least 2 thesis playing out at the moment for the AI trade, and you don’t need to be a professional trader to take part: one is the AI impact on saas (the market has been very bearish on SaaS companies, and still hasn’t corrected meaningfully), and the ai infrastructure (hardware companies + hyperscalers)
The average professional trader might not be that smart, who knows. But:
(A) I am not confident I am smarter than the average professional trader here.
(B) You don't need to win against the average professional trader: you need to win against the smartest ones. And: I'm not so sure I am smarter than the average professional trader anyway.
About your theses: my null hypothesis is that these things are already priced in.
You actually don't need to win against the smartest ones, it's not really who is on the other side of the trade. There is an interesting dynamic where the space where retail investors operate isn't really the same space as the "smart money", and because of that there are a lot of small-ish arbitrage opportunities that you can advantage based on your industry expertise (I don't mean insider trading, I mean your understanding regarding how your industry is evolving). Relying on an ETF is perfectly fine, it's way less stressful and gives you the safe-ish exposure to the market most people want. But if you want to do active trading by picking stocks (and you don't mind risking losing your money), you really don't need anything too sophisticated.
I like it how morningstar provides a graph of the stock price and morningstar's target price for the past few years, and the target price always closely follows the stock price, even as the stock price fluctuates wildly. All that genius analysis somehow concludes that the company is worth exactly what the market says it is, right up to the point that it doubles/halves in price.
I wonder with the spacex thing if this should be revised. “Smart money” have figured out what you are doing and they are at the gates.
Sorry, could you please explain?
For the record, I deliberately picked an index fund (VWRA) that's not choosy about who to admit, so SpaceX would have been in there pretty quickly no matter what shenanigans they are doing with the S&P500.
> As a retail investor, you should buy an index fund and then forget about it.
That definitely was true.
I am unsure it is still true. Index funds have taken so much of the trade volume the are becoming momentum strategies
So long as you are happy following the market wherever it goes, and if the recent past is a guide then up is the direction, then yes.
But given the nepotism and corruption in the highest reaches of USAnian society (e.g. Trump's crypto currency scams and the blatant inside dealing and rule ignoring of the Space X float) the future looks much less certain than the past
> I am unsure it is still true. Index funds have taken so much of the trade volume the are becoming momentum strategies
I don't understand how that's supposed to work?
Btw, keep in mind that index funds are typically really, really keen to lend their shares out to short sellers.
> But given the nepotism and corruption in the highest reaches of USAnian society (e.g. Trump's crypto currency scams and the blatant inside dealing and rule ignoring of the Space X float) the future looks much less certain than the past
That's a big part of why I am invested in a global index fund, not anything America specific.
The future always has risks but the question is that does an option better than index funds exist?
Depending on jurisdiction and taxation, yes. Eg in many places owner-occupied housing is favoured, and might make sense to acquire, even though otherwise it's silly: a single lumpy usually highly levered position; no diversification; multiple times your networth and with very high transaction costs.
I think there is now so much passive investing that wall street and tech bros are gaming it, and it's no longer based in reality (fundamentals). I mean Elon's deal to get SPCX short listed with NASDAQ was directly targeted at 401(k)s.
Not sure how it's all going to play out, but this ginormous increase in passive investing over the past decade or so, mainly in S&P 500, seems like a vulnerability. Small cap might be a better (non-sexy) target long term.
> I think there is now so much passive investing that wall street and tech bros are gaming it,
Yes, they are supposed to! And they are supposed to compete for the privilege. That's how index funds can add and remove stuff from the index so cheaply.
I agree that the S&P500 is not an optimal index. I picked something (VWRA) that's more diversified and less picky about who to admit.
Don’t forget there is a constant pressure for everyone to have an opinion on how this is going to end while hoping it ends tomorrow so they can be vindicated. The dotcom bust took a decade to grow and collapse. I think it is too early to make predictions with AI. I mean the sentiment here is either it will dry up the world and kill us all or transcend humanity, there’s no gray area. I don’t want to fall into the emotional sieve that seems to drive everything.
That's the financial stakes here. That's why it's all or nothing. You're spending on a level that is only justified by the bonafide machine god being ushered into existence, not productivity or coding tools (and on relatively short time horizon). So if this doesn't change the near term trajectory of humanity to a parabolic move upward there is going to be a lot of economic pain. It's not just the spending, it's that the expectations for the returns to justify them are in a relatively short period of time.
> if this doesn't change the near term trajectory of humanity to a parabolic move upward there is going to be a lot of economic pain
"trajectory of humanity to a parabolic move upward" is poorly defined here. Whether we are headed to a machine god ruled scenario or "just" incredibly powerful productivity tools, there will be a lot of economic pain for some (most) and a lot of economic gain for a few.
I've yet to a see an LLM/agent-based business plan in where scaling with an order fewer workers than before LLMs is not a central part of the value proposition.
such a business plan has not yet created economic value, being able to roll out features at rocket speed is not a huge determinant of a startup's success.
> such a business plan has not yet created economic value, being able to roll out features at rocket speed is not a huge determinant of a startup's success.
Speed of feature implementation isn't the objective. High feature throughput with labor cost reduction is.
What’s your reasoning for saying that the spending level requires that level of justification?
I think these large numbers are casually thrown about, but the real meaning is mind boggling. 1 trillion dollars is the entire US defense budget - aircraft carriers, nuclear submarines, health care, salaries, stealth fighters ect. The hidden AI debt alone is more than that https://asia.nikkei.com/business/technology/five-us-tech-gia... just for five tech giants (not to mention all the other smaller players like neoclouds)
Most people mean this to say that 1 trillion is a lot of money, but it still comes back to what you believe AI is- in hindsight, does 1 trillion dollars to build the internet sound like a lot or a little? (That is, spending 1 year of USA's defense budget to get the entire internet)
It comes back to your perception of what AI is because to people who say AI is glorified auto-complete won't believe that the money is worth it.
The AGI-pilled true believers who say it will end all money and result in a post-scarcity world believe literally any amount is justifiable.
Most people, me included, land somewhere in the middle- it seems like AI is a humanity-level sea change in technology and how computers work and serve us. It seems plausible that a few trillion is a reasonable amount.
I'm not going to make a prediction of what will happen with AI whether it will autocomplete / productivity or AGI. I will say it seems to be trending towards former than the latter just by how scaled down the promises have become over the last year (we went from curing all disease and cancer / post-scarcity to productivity and code.) The amounts being spent on this can only really justified by some paradigm shifting returns and within the timeframe investors expect. This isn't something like Apollo / Manhattan project - those were taken on by the government with public money. This is explicitly a profit making enterprise funded by markets.
1T is big in the absolute sense, but that's simply the scale these big tech companies operate at. Go back to 2024 or 2025 and you'll see as a group they are making a net income of $400B+. The scale at which these companies do anything is just staggering.
Good that you made the comparison with defense budget because it's becoming evident that AI is the new "nuclear bomb". First, if you have more advanced AI than your opponent, you just hack them and the war is over before it began. I'm simplifying obviously, but the point that AI is essential to modern warfare stands. Second, even during peacetime, you can use AI to directly influence what people think, what they do, etc. in the most literal sense of this expression. We already know that people outsource thinking, relationships, and cultural expression to LLMs, and even if that wasn't the case, you can use AI to deliver to each single person hand-crafted propaganda, not to mention the previously unimaginable opportunities to spy on people.
It would be strange if the race to wield this power wouldn't result in AI getting pumped to the moon, way beyond anything that seems reasonable.
That could all be true, but the problem is however powerful it is, it still needs to make money for the people who invested in it who are expecting a return. The stock valuations, the bonds yields - they don't care about the power of a nuclear bomb. They want to get paid. They expect to get paid. And if they don't get what they are expecting, there will be hell to pay in the economy. There's a probably a good reason the power of nuclear weapons isn't an ETF I could buy into.
By that measure it doesn’t sound like that much.
You’re talking about about an amount that is a 13% of the total US government spending, of which is 20% of the entire US GDP.
I’m not saying it’s insignificant but it’s only a few percent of the US GDP and it represents spending over several years.
I mean that's so far, it continues to grow exponentially larger with each quarter. The debt issuance for the first half looks to be crowding out US treasuries in the bond market - https://www.bloomberg.com/news/newsletters/2026-07-23/ai-deb... - that's an extremely large amount of debt. And it's still getting larger and larger each quarter.
Yeah this is going to be north of 10 trillion by the end, I would wild-ass-guess. Inflation adjusted it's larger than the manhattan project, apollo program, works progress administration, hell, it's on par with the cold war era military buildout, or a baby world war.
I would question the idea that highly industry consolidated debt is competing with risk free debt issued by the US government. Those are two very different products.
And while the quarter by quarter growth may seem astonishing it very different saying “debt levels today are alarming” versus “if this trend continues debt levels will be alarming”
So the disclosed balance sheet debt is 1.35 trillion and then the off-balance sheet debt is 1.65 trillion for a total of 3 trillion in AI debt for the 5 tech giants so far. It's multiplying every quarter and they've set investors expectations to be that this is never ending basically. But the tech giants aren't the only people spending themselves into massive debt, think of the CoreWeaves and the Nebius and the hundreds of other smaller companies. And the expectation is that there will be a near term return on all this with a healthy profit. Those five tech giants are just the tip of the iceberg in terms of the amount of debt.
While the debt numbers might look large, don't forget how much revenue is coming into these companies as well.
OpenAI gross revenue was $4B in 2024, $13B in 2025 and estimated to hit $26B in 2026. Revenue is doubling or more.
Same with Anthropic. $1B in 2024, $9B in 2025 and estimated to be $47B in 2026.
All the AI companies together are pulling in hundreds of billions of dollars per year in revenue and it's going up quickly. To me, taking on trillions in debt in order to make hundreds of billions in revenue doesn't seem crazy.
What’s weird is how emotional people get on this. I told publicly (because I was asked, not out of an obligation to have an opinion), that the prices we pay for LLMs are likely to go up because that’s what happens when the ratio of operational assets to foreign capital drops due to the capital having been turned into heat rather than operational assets. The grief I got from people, dear Lord…
I think that opinion is as reasonable as any. I feel compelled to argue against it (I even thought out the arguments in my head!) but my compulsion to have an opinion on HN is a disease, and you made a point of saying that you gave the opinion because asked.
I feel obliged to step in here to say there is a grey area where these are useful tools for some applications but not on the path to AGI.
Unfortunately the hype machine has far outstripped their capabilities so far, and the amount of money spent doesn’t look like being recouped, so somebody is going to lose money, as people lost money on the overpriced spacex ipo (overpriced because of AI).
> The dotcom bust took a decade to grow and collapse
I'm inclined to think the collapse has already started but nobody wants to see it yet.
In the last few weeks SP500 is down, kospi is down, nikkei is down, US inflation is still high and growth is softer than expected. Hyper inflated stocks (Tesla, Nvidia, SpaceX) are deflating. US bonds are at a 20 year high.
Interesting times ahead.
> there is a constant pressure for everyone to have an opinion on how this is going to end while hoping it ends tomorrow so they can be vindicated
In what bubble does this pressure exist?
How do you stay out of it all, if at all?
> Gee, AI is so complicated, how can I keep up with the landscape?
The interesting thing is: you do not need to keep up. It’s actually way easier and cheaper to wait a bit for the chaos to stabilize, then learn to use the tools. You don’t need to have been someone who experienced the whole evolution, non stop at the edge. It’s ok to let the enthusiasts discover how things work and eventually learn from them. Just like any other technology. The whole „you will be left behind“ is nonsense. If AI is the future, then it will here to stay and you can let others map the domain first
This is the way. You can jump in and out at any time and let everybody else wear the pain of working it out.
I think there's a common misconception at play here: using proprietary software thought of as a skill
Must be a mix of decades of corporate branding in software and the erosion of trust in the labor market coupled with the rise in noise from various sources...
Skills are what they always have been. Logical thinking, good memory, long attention span, critical thinking, communication (the ancient Mediterranean, the culture which forged the currently dominant "West", valued 'rhetoric' almost above everything else besides physical prowess)
Being well connected, having a sharp mind, cultivating deep knowledge of science, philosophy, language, mathematics. These things are timeless. Using the current brand of software well is a moot point. Software can only become dominant if it falls in line with mainstream design. Learn design, you can intuit how to use relevant software and the badly design tools will be weeded out naturally.
In a nutshell, exercise body and mind. Be strong, cultivate strong relationships. Everything else is just background noise.
I was able to create a custom index based on the top 500 that I stripped the big AI stocks from (shovels too). Then I added decent chunks of international, small cap, and treasury ETFs to it.
I have no illusions that I can time a bubble, but I'm hopeful I'm at least partially shielded, and most importantly I feel better about ignoring wall street again.
Instead of starting with the top 500 American stocks, and then adding international and small caps, you can start with a global all-market stock index--and then remove AI from that.
Yeah I couldn't figure that out with Questrade (Canada). It's a pretty new feature, but I think it's great, so I hope they expand their baseline indexes.
I was considering writing a tool that simply follows any index you choose with a .toml of simple config options, like which stocks to exclude, potential fixed locks for specific stocks (or maybe upper and lower percentage of portfolio settings), a hard per stock cap (say AAPL at 3%), and drift threshold. Something you just run once a day and it spits out your buy / sell orders. Seems like this is something brokerages are already offering in some variation though, and I'm not sure what, if any, API access looks like, or export / import options.
I'm in Singapore. My money is in VWRA (without bothering to remove AI companies).
Your idea for the tool sounds interesting. I suspect even just copy-and-pasting the paragraph you wrote here into your favourite AI programming agent would get you pretty close to a prototype you can play around with. At least in terms of 'spit out buy / sell orders' and leaving out the API integration.
Yes, it's a very simple concept IMO. Without API access or at least CSV import / export integration w/ a brokerage for automation I don't think I'd use it. I could have an agent use the Web UI on my behalf, but honestly, that feels like lighting tokens / gas on fire.
So at least IBKR has some MCP (or so) integration.
If you are not trying to be current up to the minute, you can get stock information from eg Yahoo or elsewhere on the open web.
You are perhaps not as diversified as you think in this top heavy market.
https://stockanalysis.com/quote/lon/VWRA/holdings/
~ 5% Nvidia as biggest holding and 20% in US listed tech companies (most of which are heavily invested in AI), over 60% in the US market, so this ticker is very similar to investing in the US market alone.
Also when a bubble like this deflates it hits almost everything so it is very hard to avoid, but world indexes are particularly exposed.
You are right about the numbers. But wrong about what I am thinking: I'm aware, and it's a deliberate choice on my part to allocate in proportion to market cap.
>I can time a bubble
What do you mean? Selling everything before this bubble pops?
Oh, if you could time a bubble, you wouldn't just sell everything: you'd sell more than everything. Ie you'd sell short.
Talk about taking a quote out of context...
> taking a quote
What exactly do you mean here? Like quoting it the right way?
Warren Buffet way
Revolutionary technology + massive adoption ≠ good investment
Investors have poured money into a bottomless pit, attracted by the growth and glamour of the industry. The airline industry since its birth has had a collective net loss, in aggregate, despite moving hundreds of millions of people.
Commodity Product, no switching costs. Infinite competition
The airline industry since its birth has had a collective net loss, in aggregate, despite moving hundreds of millions of people.
The industrialisation essentially socializes the cost across a lot more people though, so even though it doesn't make a profit it does mean people can have air travel without it costing millions per flight for the few people who can afford it. Essentially the economies of scale from having lots of flights isn't enough to make it profitable but they are enough to make it affordable.
There's no spare money to extract from the airline industry but it's still very useful. The same could be true for AI in the long term.
Sometimes the goal of an industry is to exist rather than to make a profit, because the benefit to society is more important than profit. People don't like that though so they do a bit of creative accounting or head-in-the-sand denial around it.
> There's no spare money to extract from the airline industry but it's still very useful. The same could be true for AI in the long term.
Of course it could, let’s start with making the models open weight and entirely open source. Fully publicly owned and not shaped to maximise profits for the shareholders.
Oh wait, Scam Altman entered the chat and turned a non-profit lab into the next biggest IPO vehicle the world has ever seen.
OpenAI launched as a nonprofit research institution. Its announcement explicitly said it wanted to pursue AI “unconstrained by a need to generate financial return,” produce value for everyone rather than shareholders, publish research and share patents broadly.
in the case of airplanes the only thing thats the private market is the planes and the ticket, the entire system of airports, safety, navigation is state subsidized and when the market fails it gets bailed out. the oil is subsidized by constant warfare. it's just an illusion for reganomics so a few rich ppl can make a buck off of a public utility.
How does constant warfare subsidise oil? In case you haven't noticed: both the latest US-vs-Iran war and Russia-vs-Ukraine war have made oil and gas a lot more expensive than the peaceful counterfactual.
> The airline industry since its birth has had a collective net loss
True. But it has added enormous benefits to many other parts of the economy.
Airlines do not capture that value.
That is where the AI companies are. Adding value they cannot capture
Yep, and similar to airlines, the real money is in suppliers
It isn't a commodity product in my opinion. Far from it. I think it will ultimately be a monopoly or duopoly for SOTA. The mid to low end is commodity, yes. But SOTA models are not commodities.
The number of competitors for SOTA drops by a few every year. The winners make more money, get more revenue, buy more compute, train better model with compute, buy best talent, and the cycle goes.
I think it's easier to fall behind and never catch back up than people think. One disastrous training run can leave a lab months to a year behind. For example, Meta's disastrous LLAMA 4 models. Meta is lucky to have their ads business as a funding source. However, Anthropic's revenue is growing so fast, that ability to use ads as a funding source to stay in the race may not last much longer for Meta.
To me, SOTA LLM training is very much like new chip fab nodes. One disastrous node can put you behind for many years or forever. The cost to build the next chip node doubles every every 4 years (Rock's law). The cost to train the next SOTA model likely has some similar power law which means over time, it's too costly for losers to keep up. The only reason TSMC isn't a defacto monopoly for advanced chip nodes is strictly due to geopolitics.
But who needs SOTA models, really? It was necessary 10 months ago, but now?
All things equal, let's say your SaaS startup uses GPT 5.0 (release 10 months ago) and my business uses Fable 5. We have the same business goals, same talent level, same strategies. I think the chance of my business winning against yours is higher.
I can't prove it. It's just my opinion.
It's easy to agree to that, but you're disregarding that the resources you spend on the stronger model could be allocated elsewhere. Conversely, you're assuming that spending more on AI will always yield better results and be worth it, compared to spending the money on other things.
This might actually still hold true now, or or at least many actors in the market behave that way. But I'm not so sure there isn't a cliff to that effect. At some point, if SOTA models remain expensive, it'll turn into a market advantage to figure out how to get things done without depending on the most expensive tooling available.
Similar scenario, different phrasing: if your company relies on overqualified workers to deliver 100% quality, the market may still decide that it's fine to go with 90% quality for 50% the price.
@orwin has claimed that SOTA LLMs have already hit that diminishing return where spending more money on a SOTA LLM today does not add more value than a non-SOTA LLM (assuming high value tasks).
I never said there will never be a diminishing return. I'm challenging the statement that we've already hit.
Note: We're still scaling chip nodes. It's still worth it for TSMC and chip design companies to invest hundreds of billions into every new chip node every 2-3 years. This is after decades of scaling already.
I was thinking about Opus 4.5 when i talked about 10 month ago, but maybe i'm confusing the date, it might have been 9 month.
Opus 4.5 vs Fable 5? i think the output difference will be negligible, and the opportunity cost of paying an order of magnitude more will actually benefit my business.
I'd guess it depends on the type of business. If it's some genuinely deep technical space, the model would give an edge but even then I think luck would be a significant factor. In a monte carlo of such scenarios, the business with the stronger model might win 6 out of 10 times, but it's no sure thing between the two of us. If we were comparing two businesses building Yet Another Generic CRUD, I would guess it's closer... perhaps even a net-negative to spend money on Fable versus marketing.
Your business is spending way, way, way more for compute. How can you ignore the economic of your business? That’s pretty much what a business is about
Ceteris paribus, all other things are never equal.
That’s nonsense and saying “nah-ah” with Latin won’t improve your argument.
If we couldn’t isolate a variable we would never be able to argue.
Using a better model is an advantage even if only for the coders. There are a million ways to turn that into profit, both proper and not so proper but that’s the beauty of ceteris paribus: the other factors do not matter now.
> There are a million ways to turn that into profit
Mind pointing where that profit for companies consuming AI is? I don’t mean hypotheticals. Where are the proof that current AI contributes positively to ROI?
I think the chance of success of the GPT 5.0 startup is higher since they aren’t just gonna be relying on Fable to vibe code some slop saas.
There is literally no moat…
Then start your company and go compete against OpenAI and Anthropic then.
Heck, just take Kimi K3 and start your training run from that.
thank you for that great contribution to the discussion...
Key reports to understand the root problem (no ROI):
- Gen AI: Too Much Spend, Too Little Benefit?: https://www.goldmansachs.com/insights/top-of-mind/gen-ai-too... (Goldman Sachs)
- AI’s $600 Billion Question: https://sequoiacap.com/article/ais-600b-question/ (Sequoia Capital)
- The Simple Macroeconomics of AI: https://www.nber.org/system/files/working_papers/w32487/w324... (MIT / Daron Acemoglu)
How can people in Hacker News still doubt AI's benefit when they are seeing in front of their eyes every technical profession getting disrupted to oblivion in the last year. Just ask basically any software engineer how much their profession has changed over the last 12 months
Obviously there is risk, but can't we really extrapolate the AI gains forward and just see how big it's ahead to become?
> every technical profession getting disrupted to oblivion
Where is this disruption? The longer we go, the more people report that the supposed net-gain of easily 100s of percents is not visible.
I do strongly believe "It's just a tool" - A powerful one, but not one like the invention of the steam machine.
All I see is a flattening of the technical curve. Which is great, but the number of people who want to download an app is still the same. So all you have is 100,000 apps with no users instead of 10,000 apps with no users
You increased the amount of code written by 10x but unless there’s a 10x increase in demand, its worth nothing
https://www.frank.computer/blog/2025/05/just-a-tool.html
Tell me you weren't around during the dot com boom without etc, etc.
The benefit is neither here nor there - it's whether the borrowed money will ever be repaid on the lenders' terms.
I don't doubt that AI has benefits, but I do doubt that the major AI providers will be able to make back their investments. They've spent trillions of dollars, and yet they've barely created a moat. We're seeing open weight models being released that are only months behind them, that can run for way cheaper. This makes the future of OpenAI and Anthropic suddenly look rather bleak.
A simple analogy: If you have kids, you love them and want to give them whatever makes them happy. But on the other hand, you run a household, you pay for bills, healthcare, heating, education, and heavy overhead. You must keep things under control. You don't hand a blank check to an immature child who doesn't even know how to manage that money yet, right? So, even if your child wants to push forward at an extraordinary pace, you have to keep a level head, manage spending, and ensure everything doesn't end in ruin.
That’s the point: making growth sustainable over time.
Not seeing any significant disruption showing up in revenue etc from inside the industry, there’s growing skepticism about these tools and the claims made for them. They can be useful but are not world changing or actually replacing jobs.
We’ll see how they develop but so far they are not capable of operating independently.
> How can people in Hacker News still doubt AI's benefit
The internet and railroads were highly beneficial, still crashed the economy.
I know many companies are spending quite a bit of money, I don’t know if it bears out that the increased spend has resulted in increased profits, even if there has been some increase in productivity. I think this is the tough situation many orgs are facing right now, drastic adoption without material economic gains.
I use AI a lot but my final output isn’t drastically different
Writing a lot of code doesn’t mean much when the moat was never “writing a lot of code”
disruption != benefit.
I was called quite disruptive in class when I was young. I'm sure my teachers never meant it as a compliment.
How can people not trust in anecdotes and vibes while avoiding studies, do you mean?
Isn't that the point here? That everyone thinks massive disruption is happening and everyone is 100xing their productivity, but it's not actually showing up in the numbers anywhere?
> How can people in Hacker News still doubt AI's benefit
Because what many of us are seeing is meaningless “productivity” improvements.
If at the end of the day you don’t have more users paying for your product or the same users and paying more, then what’s the point of being more productive?
Stop projecting.
It’s just coding. It’s not every technical profession at all.
People aren't doubting the benefit.
Lenders are doubting their return. People's benefits have nothing to do with it. The benefits would go in a minute, if doing so yielded a better return.
People are being a little unfair to you here, I think. I think there no chance of AI not being by far the biggest technological shift in our lifetimes. BUT that doesn’t mean any of the current companies leading the charge have sustainable business models, or that the current financing around it makes sense. Other commenters have pointed out both the railroads and dot-com boom as analogies, which holds up well. Generative AI is here to stay, but that in no way means that Anthropic and OpenAI are
The Internet was hugely disruptive. The dot-com bubble burst and tech stocks flatlined for years afterward.
The headline and the link have nothing to do with each other.
Am I too dumb to read or this thread is heavily botted? Why most top comments are unrelated to the linked site (which is pretty obscured already)?
Make the most of your heavily subsidised $20 / $200 subscriptions while the credit spreads allow it.
I’m not sure that’s quite the right framing. If Anthropic goes bust, Fable persists as an asset that can be run by someone who didn’t have to pay to develop it, probably profitably, and probably in a way that gets cheaper over time. The debt pony show is paying for the next model.
> Grey Swans: risks that were in the data but overlooked or dismissed because few had synthesized the signals into a coherent picture.
Directly conflicts with
> Alert and Critical signals represent readings that have historically been associated with meaningful financial stress.
These are all pretty standard things to track and are regularly (and publicly!)
Not saying we’re not in a bubble or near/far from it popping, but these metrics aren’t going to precisely tell you _when_, which is pretty much the only thing that matters.
Speaking of AI stock, what has happened to NOK?
The link is broken
I am sorry, but this site is vibecoded beyond my comprehension. What is it trying to say?
Everything has been running on borrowed money since the 17th century.
As a side note. All money is borrowed. That's how money gets created! Short explainer video by the great late David Graeber
https://youtu.be/LxJW7hl8oqM?is=IjdyHwZchaiMHk4C
First, and obviously, the article is talking about capital when they say 'money'. Not all capital is borrowed.
Second, not all money is created via borrowing (but the vast majority is!)
And the YouTube video you linked to is very confused even about the money that is created via borrowing.
Government debt is not required to create money. The Bank of Japan bought stock ETFs to get 'freshly printed' money into circulation. ('Freshly printed' in scare quotes, because these days it's just entries in a database.) Another example: Singapore's central bank (MAS) does not use Singapore government debt to create Singapore dollars; I'm not even quite sure they would even be allowed to.
You can say that money itself is a debt of the central bank; and that's sort-of true, but it's not what David Graeber talks about.
A bit of a pedantic last point: silver coins or bitcoin also require no borrowing to create. Silver coins have been used as money, bitcoin could conceivably be used as money. (There are other problems with these options, but that's besides the narrow point.)
Totally depends on what you define money as.
If "money" is the some function of all outstanding credit, then yes, it is created (mostly) by bank lending
If you define money as a web of trust then it is mostly created by those that create the rules. The state
“None of this yet resembles a credit cycle turning, and the signals that would show one are quiet… What the data describes is closer to the opposite of a contraction: a credit expansion absorbing record supply and charging progressively more for it. The pricing runs in a ladder.
…
Where the ladder breaks is at the bottom. CCC and Lower Option Adjusted Spread signal rose twenty-two points over thirty days to 88 and sits at Critical, markedly higher than the Investment Grade or High Yield spread signal levels, and AI paper does not price at CCC. Private Credit Stress sits at 96, up thirty-one points, consistent with reporting data center financing has moved toward private credit and off-balance-sheet structures where the ultimate holder is harder to identify. What separates an expansion from a contraction is not the level of spreads but whether new issuance keeps clearing. An expansion growing more expensive still places its paper; a contraction is when deals stop pricing at any spread. The market is still clearing. It is clearing at a price that has moved in one direction.“
The real cost isn't the inference bill, it's the senior engineer hours spent debugging subtle hallucinations that slip past naive test suites. We stopped bleeding money on trivial boilerplate by implementing model routing through FlintAPI, ensuring only complex logic actually hits the expensive frontier endpoints. Treating LLMs as a monolithic tool instead of a tiered pipeline is what quietly destroys your margins.
you won't get debt if you don't have assets that can be repossessed, so having debt means these AI companies have assets: that's a strong thing, not a weak thing. interest rates are what they are, and they go up and down for reasons exogenous to your industry; debt regardless of interest is always "cheaper" than equity, and the shareholders expect to make their money from equity, paying interest on debt as a type of impedance matching and cost of keeping more equity.
so everything is going according to plan, and nobody knows the future, and predicting collpses has never been a profitable business.
I didn't have to read past the first few confusing contorted and convoluted paragraps of this article to decide to come over here and explain it, this is all straightforward corporate finance 102 and the article is fluff
I agree with the general sentiment, but I feel like it is also a bit reductive. Assets in this space are near impossible to evaluate and can fluctuate in value greatly based on other actors. In a hypothetical scenario where, say, google releases a new frontier model that somehow leapfrogs the competition by 5 months all of a sudden the value of the Asset of Fable 5 and GPT 5.6 might completely crater.
>you won't get debt if you don't have assets that can be repossessed, so having debt means these AI companies have assets
That's the ideal scenario, but you can also have immense debt because someone once thought you had assets.
Yes, they have assets: GPUs sitting in datacenters, and data.
Question is: is that worth enough to cover the debt after the market crashed?
Don’t they all have mostly the same data, with a small / negligible delta between each other?
I'd argue that data in this case is more like the actual models they use, their codebase and their engineering talent. Not deep enough in the sauce to say one way or another how big the realistic delta between companies is though.
Maybe so. That leaves just the hardware. Deprecates badly. Maybe gamers will take it off their hands gladly?
I would be delighted to see a glut of compute. I'm not optimistic, though.
GPUs have a five year lifespan before they become obsolete and start experiencing reliability issues. We're already 1-2 years into that five year lifespan.
The payback time for a GPU running 24/7 inference is ridiculously short though. As little as 6 months according to some calculations. Most of that 5 year lifespan it will be earning well in excess of its replacement cost.
I would imagine Anthropic et al. are largely leasing land/buildings, so as the other commenter said… must be the server racks that are acting as collateral (if anything). Generally enterprise hardware depreciates very harshly. I’m used to paying $10 for Intel Xeons that once retailed for over $5,000. I expect to pick up some NVIDIA Blackwell 6000s for $100 each someday.
Yep, a friend recently told me that he remembers working somewhere that gave away old empty server racks - they were unnecessary, and expensive to store, so why keep them?
We are in odd times however - I for one am sitting on paper profits on the consumer gpu I bought 2 years ago. If anyone goes down before the supply side is fixed - the first to fall will probably be able to liquidate their gpus at a profit.
> the first to fall will probably be able to liquidate their gpus at a profit
Meta and xAI announcing they are leasing out capacity is a version of this already happening.
I remember when Amazon was going to go broke every year for over a decade.
Until they didn't.
That wasn't what it seemed like at the time. Amazon didn't post profits, sure, but they sure as hell weren't a giant money suck either, they didn't need billions in financing to run their business. There were a lot of Amazon bears, but they were concerned about the high valuation, not about them going broke (since even the most pessimistic bear can read a cashflow statement).
That’s because they were reinvesting the profits. I think they had given a profitable quarter just to show that they could do it.
And I'm sure Anthropic would be immensely profitable if they stopped investing their inference profits into training newer models.
And then everyone would stop using their inference as soon as a better model for a reasonable price came out.
The R&D expenditure is a critical requirement for the inference profits, to the point where we should probably lump their financials together, at which point is definitely not profitable.
What will it look like when R&D plateaus (and yes it definitely will, but it could take a while), investment falls, and a few main competitors remain in the music chairs?
It's very difficult to predict. The inference profits we are seeing the profits of a company that is temporarily ahead, but the revenue will level-out in a more stable market, depending on how many survived. It's also hard to tell where the costs will be at the end of the game, with constant efficiency optimisation mixed with cost increases for higher intelligence.
I think it will be quite similar to the semiconductor industry, where, yes there are some key monopolies, but they are not the initial big players, and none of it is actually very profitable; while the real profits are reaped by those that make popular consumer products based on the foundational tech. I guess the main difference is that OpenAI and specially Anthropic have been quite effective at directly tapping into the consumer market rather than remaining technology providers.
It's never that simple, that's not the only possible endgame.
We have seen plenty of examples in other industries where you can never really stop investing a ton on R&D with diminishing returns (like in semiconductors or pharma), because the moment you stop newcomers overtake you.
Or the whole thing becomes a commodity with lots of competitors, where technological advantage is overtaken by marketing as the dominant force.
Or you really are left as the only player alive, but you realise that the market cannot absorb higher prices for your product by then, they prefer just not to buy it. Perhaps you are the only player alive because the business has become so low-margin that everyone else has abandoned it intentionally.
That Silicon Valley pitch you are echoing rarely works out as advertised, even for the winners.
I think you're proving my point. Eventually, R&D heavy industries almost always become a monopoly or duopoly. Small/losing players can't keep up and drop out or acquired.
Check their R&D expenditures and profit margins they are quite poor.
Their revenue has always been sustained by the fact that their technology needs to be constantly replaced because it keeps getting better. The moment it stops getting better, the replacement rates plummet and so do their revenues. It's also really not that hard to compete with them when they get complacent.
I'm not counting Nvidia because they don't produce semiconductors themselves, they are a different kind of business.
They are indeed an example of those that cater to consumers and/or build popular products based on foundational tech from others, like Apple or Sony, which do tend to be quite profitable.
But the actual deep-tech semiconductor firms? They may be critical to the world economy, but they don't actually make that much money comparatively. In many cases they are not real monopolies, it's just that no one else wanted to continue investing in a shitty business model. Only the likes of TSMC and Samsung were okay in playing the low-margin game, but most US players left the board.
I believe AI has a lot of the same characteristics.
For how long though? If Amazon never built AWS the core business conceivably would still be around today, if Anthropic stopped providing new models two years ago no one would care about them now.
An LLM model depreciates to $0 in a year or two.
An Amazon fulfilment warehouse depreciates to $0 over 30 years.
One type of investment is different to the other.
How much does that warehouse make vs that LLM model? IE. How much has Opus 4.5 made for Anthropic?
Amazon had a close call around the .com crash as capital markets froze, but they were not going broke every year. They were purposely (and rather famously in business circles) investing every dollar made in order to grow the business. It was clear early on the original business worked.
Amazon also added/pivoted to AWS, which is where a huge part of its value comes from today.
I remember when hundreds of dotcom companies were going to go broke, and they all did.
Not sure what your point is.