As long as this debt does not make it into life insurance and pension funds, we are fine. The trouble is that private credit is taking control of some life insurance companies and off-loads this debt to these. When these fail, it will become everyone's problem.
> Risks to financial stability may also stem from entities with particularly high exposure to private credit markets, such as insurers influenced by private equity firms and certain groups of pension funds. The assets of private‐equity‐controlled insurers have grown significantly in recent years, with these entities owning significantly more exposure to less‐liquid investments than other insurers
Couldn't it be a problem given the concentration of the S&P in these companies?
At this point these companies make up a huge portion of 401k's for a huge chunk of Americans. How would it affect retirees if they dropped 40-50%, likely taking the market with them?
I suggest looking into “EQL”, or better yet, just replicating its index by taking a position in the 11 XL* sector funds from SPDR, allocating equal weighting to each. One will end up with one’s equities equal weighted by sector and with plenty of large cap exposure, as opposed to the pronounced mid-cap tilt found in whole market equal-weight strategies.
Personally, I drop the financial sector entirely (Thomistic prohibitions on usury) which leaves an even 10 funds which is easy to allocate mentally and in practice. For example, assuming a 60/40 allocation where one is holding the lion’s share in equities and the remainder in bonds (I substitute with a combination of gold, crypto, cash, and Swiss Franc here), one would allocate as follows:
(Note that XLF is consciously not taken as a position here, decide if it’s right for you. The Mortgate REITs which would make XLRE problematic are in XLF per the sector selection rules)
The remaining 40% is bonded debt if you are fine with usury, or some sort of asset negatively or neutrally correlated to equities.
It’s an interesting thought. The growth is so extreme that if the S&P 500 fell 50% today it would reach levels last seen in 2022. Given that the timespan is so short, I’m honestly not sure it would be as bad for 401ks as people expect unless all of your investment was concentrated in the last 4 years.
I suppose it’s worse if your calculation is, “I’ll retire when my 401k hits $X absolute value,” but I think most people just retire at a certain age instead with risk spread across decades.
During the dot-com crisis. Nasdaq fell around 78% from its peak and S&P by around 49% so it isn't unprecedented (ironically has both aspects of being both tech and are within the same time-era)
It created an actual recession albeit thankfully short one for the case of dotcom (sadly not for 2007) and a really recessionary environment which causes unemployment and just straight up fear and panic.
I do understand what you are talking about and overall in long term, perhaps things flatten out but atleast speaking financially so, its better to be on a smooth sailing road rather than insane ups and downs with retirement money if preferable.
> I think most people just retire at a certain age instead with risk spread across decades.
The issue in my opinion is with people near that certain age you mention and who retire in the time during boom just before bust. They would then get the 50% hit on their savings instantly with an recession/inflation/unemployment environment which in my opinion might be genuinely devastating.
(supposing that they had their investments in stocks, I wouldn't consider that any retiree would have all their money in stocks but there have been some other comments which show a sizable amount, @kipchak's comment shows 50% stock for retirement. so a 50% shock on top of that could lead to a wipe out of 25% of your retirement fund which is honestly still pretty crazy.)
> (supposing that they had their investments in stocks, I wouldn't consider that any retiree would have all their money in stocks but there have been some other comments which show a sizable amount, @kipchak's comment shows 50% stock for retirement. so a 50% shock on top of that could lead to a wipe out of 25% of your retirement fund which is honestly still pretty crazy.)
What do you think the cost would be for protective puts per $1M of equity exposure of a portfolio on a monthly basis? "Bubble Insurance" if you will. Going 100% bonds is simply intolerable for the time window for most retirees, but an equity wipeout of this magnitude is equally intolerable.
NVidia makes up 7.5% of the SP500. If it lost 50%, it would be a 3% loss for the index. The concentration is bad, but it would not cause a drop of 50% retirement funds by itself. If you take an all world index, it's even less.
Still, if NVidia lost 50% of their market share, we would probably see a big collapse of the stock market.
EDIT: to note, the top ten companies in SP500 make up an unprecedented concentration but they're not "mostly AI".
I don’t think concentration risk is itself overly concerning. The nature of a market cap weighted index means it will always be heavy on whatever is currently trending. You’ll certainly be hurting if your plan is to retire at the top of the market with just enough, as the inevitable downturn will hammer your portfolio down into not enough. So invest until you have enough to handle volatility or a lost decade with a dip and slow recovery.
retirees arent suppose to have their active retirement funds in stocks dude. Any financial advisor with a brain would not make such a ridiculous asset allocation error.
Regardless of whether it's a good idea, it absolutely happens and as a result would impact retirees, both in individually managed accounts and target date funds. For example here 70+ are 45% equity.[1] TROW retirement 2020 funds are about 50% stock, for example, and only decrease to a floor of 30%.
For those who stick to a meaningful asset allocation (e.g. 60/40, 80/20, etc), this does not pose significant problem -- they would not be buying much stock in the last 3 years. Instead, they would be buying mostly fixed-income. Probably mostly in 401k/IRA accounts.
The problem some have pointed out is that these companies are such a huge portion of the market right now.
The sound advice for the past decades has been, just invest in a low-cost ETF tracking the S&P instead of picking stocks to minimize risk and invest in the market broadly.
So a huge number of people have done that, believing they're diversified, while tech makes up 40% of the index.
Yes you could sell your S&P and find things to invest in least likely to be impacted by a potential bubble, but your average 9-5'er with automated contributions to their 401k is probably not sophisticated enough to do that.
And that's assuming only these companies would be affected if there was a massive draw-down in tech/AI related stocks. We haven't really seen a situation like this before, so it's not easy to predict what effects there might be in the broader economy.
no one that needs to rely on their investments for their actual retirement still has them in equities. theres a reason target date funds automatically adjust asset allocation as it nears its target date. you should be in majority bonds and cds well before your actual retirement date.
The employer selects a financial company to manage the 401k. When you switch jobs, you can roll the 401k from the previous employer into the new one, or into an IRA (Individual Retirement Account).
Usually the financial services company will offer several options: more aggressive/high risk, or less aggressive/lower risk. Most people will just go with whatever is the default option.
So much of the American S&P 500 is dominated by handful of companies that the risk is not that easy to avoid. If or when the AI bubble pops, it's going to take down a lot of the economy with it. You can direct your retirement savings into the lowest yield/lowest risk assets offered by the firm, but you'll forego whatever growth happens in the mean time.
Will the bubble pop next week? Next month? Next year? Who knows. Timing the market is incredibly difficult.
There's a famous quote, attributed (perhaps apocryphally) to John Maynard Keynes: "The market can remain irrational longer than you can remain solvent."
The whole idea of a pension fund is that you don't need to time the system it is the system.
Like my country pension scheme. It went through ups and downs for a hundred years but has always come on top. All you need is a long horizon and a trillion dollars and you basically can't lose.
> When these fail, it will become everyone's problem.
Debt is senior to equity. For private credit to start taking haircuts, the equity has to have already gone to zero. At that point, this will already have been everyone's problem for some time.
Are you suggesting that holding private credit assets is relatively risk free?
Equity is a risky asset, so equity being wiped out should not be a surprise to anyone, but life insurance and pension funds failures is indeed a public problem.
There's a wide spectrum of 'private credit', and it varies from low-risk to quite high risk, depending on the debtor. In the case of "AI Companies", their debt is low-risk, but many are creating special-purpose-entities which will build and own some or all of their newer datacenters. Those datacenter companies are issuing a great deal of somewhat risky debt; with the exact level of risk depending on the off-take agreement.
You probably meant to say that practically, high leverage tends to leak into companies of public interest. For example, when high net worth individuals start trimming their private credit holdings, which eventually end up with insurers. That is why the regulators have to watch carefully that it does not happen.
> As long as this debt does not make it into life insurance and pension funds, we are fine.
I think for small to even large numbers you are correct, but given how yuge this debt amount is a broad-based default will probably cause a contagion. I will not predict how far and wide.
You say this as though every company doesn't take on debt, and all debt isn't a risk. I'm sure you have some much riskier debt than ChatGPT already in your portfolio, and interest rates are adjusted by relative as judged by the market. I'm sure some debt will fail, but certainly all of it won't, and while anything could cause a market crash saying "when these fail" holds a lot of incorrect assumptions.
Do they? Is a company with $200 billion annual revenue and earnings (EBITDA) of $100 billion having $420 billion of off-balance-sheet debt really staggering?
In many other industries that would be a perfectly normal amount of debt to have. It's only unusual because we are used to tech companies having so much cash on hand they don't know where to put it
These companies have valuations reflecting a debt light business. At a minimum, 420 billion in debt is enough to change the stock price by 10-20%. If the company plans to add another 400 billion in debt you need to give it the side eye.
If 50 billion in revenue is from other companies debt spending… then You have a problem.
> These companies have valuations reflecting a debt light business.
Sorry, but this doesn’t make sense. The valuations of these companies reflect their growth.
In finance there’s nothing inherently virtuous about a “debt-light business”. It’s all an allocation decision based on how you expect to grow relative the cost of that growth.
Try and reframe it: are cash-heavy businesses given a premium?
My point was more of an exercise to point out that finance is about mutating resources. A lot of cash can be a good thing or a bad thing. Same for debt. There’s nothing inherently bad about levels.
There are many reasons to use subsidiaries for things like this, like to invite outside investment, ringfence risk, cede operational risk, and many more.
This is all like CFO 101 type stuff, and not nefarious. I find it amusing that people assume the worst for things they understand little about, rather than trying to learn.
Maybe the best way I can explain it to the programming crowd is this: imagine how ridiculous it would sound if outsiders were saying that Google was on the verge of collapse because its codebase has billions of lines of code.
> There are many reasons to use subsidiaries for things like this, like to invite outside investment, ringfence risk, cede operational risk, and many more.
Agreed, there are many valid reasons to have subsidiaries of course.
The issue is rather with the fact that we are having the assumption that the threat is outside rather than inside and so systems with mechanisms to be less transparent are far more prone to this risk.
It has been academically shown that most corporate/ white paper scams aren't done from outside but rather from inside the company itself through genuine structures and incentives which go wild. (Something shockingly visible in AI space), Enron's example also comes to my mind.
The best way I can explain it to the programming crowd is this: Imagine how ridiculous it would sound if you are ranked with how many lines of code you ship and how much token you would spend and so we end up with tokenmaxxing and hearing stories about people literally burning tokens in innovative ways because they want to get on top of a leaderboard. Oh wait, it is already happening or has happened.
Generally speaking, It is preferable to be transparent with debt and other things rather than not especially so for long term because sooner rather than later you might get caught. Obviously if there is some stuff which prefers from ringfencing risk then sure.
Also as I spoke of Enron, but the exact structure was used by Enron as well as @fzeroracer discusses in their comment[0] so it might be a genuine question.
"There are many reasons to use subsidiaries for things like this, like to invite outside investment, ringfence risk, cede operational risk,..."
In other words, they are intentionally deceiving investors and hiding the risk from them. That does not sound like CFO 101, it sounds like fraud. But if grift is your business, I guess those are as valid reasons as any.
"Companies disclose such future debt not in their balance sheets, but in annotations to their quarterly financial statements. This is a legitimate practice under accounting rules, but may make it difficult for retail investors to recognize risks."
"Today's AI industry is partly supported by demand generated by circular investment. Nvidia and tech giants invest in data center operators and AI companies, with that money then turning into GPU and cloud usage fees. Actual demand is difficult to see, increasing the likelihood of over investment in data centers."
Retail investors shouldn’t be investing in individual stocks outside of industries they understand well. Following GAAP is the definition of not hiding the obligations.
Maybe GAAP should be changed? Professional investors have the time to comb through the small print. These off balance sheet activities seem designed to bamboozle the small investor.
>but may make it difficult for retail investors to recognize risks
Ok, so just to be clear: institutional investors are (a) the ones investing the large proportion of capital in these companies and (b) are well equipped to decipher financial statements. The idea that any significant amount of retail investors have even seen a financial statement, let alone is making decisions based on their analysis of a financial statement, is laughable. And, even then, if someone is putting in that effort, then presumably they're not going to get tripped up by a legitimate practice that they ought to specifically be looking for given the context.
And all of that doesn't even take into account that every article discussing the financials of AI firms over the last half a decade have been pointing out these dynamics. We're literally in a thread discussing this exact dynamic. Retail investors are certainly far more likely to make investing decisions based on these kinds of articles and threads than they are based solely on independent financial statement analysis that they're conducting. At a minimum, I think anybody taking any of this seriously has gotten the hint by now.
If it comes out that these firms are committing straight-up fraud, then there will be a lot more to discuss. But, as of now, the sentiment is that these firms are behaving perfectly legitimately, just abnormally and maybe irresponsibly compared to their historical context. If an investor isn't equipped to handle this kind of analysis under these circumstances, then I'm not going to feel too bad if they lose their money "investing" when they're really just gambling.
they aren't trying to hide anything, those are accounting rules that are applied to the letter. I'm feeling like I'm taking crazy pills whenever I see this stuff about AI, your hate boner for a specific technology shouldn't trigger you saying things that are provably untrue.
Are they really "trying to hide" this debt? I think it's pretty common knowledge that a lot of these companies are using debt/bonds for funding. The debt not showing up where the author wants is a reporting formality not an attempt to hide it.
I think the point is that it’s not showing up on the standard financial filings. If you were to pull the annual reports for these companies, you wouldn’t see it. That doesn’t mean it’s impossible to find it. Obviously, it is otherwise the article wouldn’t have been written. But you’re going to have to go the extra mile. To be clear, none of this is illegal. It’s just covered in the advanced CFO accounting class.
It's not hidden at all. Financial blogs very accessible to laymen like Matt Levine's Money Stuff have talked about this structure months ago. If you are an investor and surprised by this news you weren't sufficiently prepared and shouldn't have been investing in the first place.
Take-or-pay contracts appear as "contractual commitments" in 10-K. They are not hidden. That's the way they are reported in all industries where take-or-pay contracts exist. There's nothing nefarious about it.
If it didn't matter, why would they bother jumping through hoops to keep the debt off their balance sheet?
In the run-up to 2008 a big factor in the bubble forming was that poor quality loans were packaged in a way to hide the risk in those investments. I'm not expert enough in finance to know if it's the case now, but we do know that clever accounting to hide debt can lead to the incorrect valuation of assets, potentially leading to financial ruin.
They're not jumping through any hoops, I think they're simply complying with reporting requirements. It's not on their balance sheet because being recorded as strait debt would itself be misleading. My understanding is that these sort of off-balance sheet "debt" is mostly in the form of deal terms that may or may not be expressed at some point in the future.
An analogy that comes to mind is when companies used to book future sales in the present. They got in trouble for this and is now forbidden. I recall reading that one deal had terms that transferred assets if certain conditions were not met. If terms-based debt should be booked now, then terms-based assets should as well. This stuff makes my head hurt.
Either way, as long as it's not hidden (and it's not for the public companies), then it's fine.
You couldn't just characterize them that way, they are using literally the exact same setup that Enron did in order to hide the massive amount of debt and liabilities they had that eventually sunk the company. People jumping in to defend this as being totally legit or not-fraud seem quite insane to me: companies should not be trying to hide these things to pump up their stock prices or to entice investors that otherwise might not see it.
One of Meta's SPVs building a data centre for them, Meta own only 20% of it; 80% is owned by other investors. That's the issue here; the market thinks that only these handful of money-go-round FAANGs/Mag7 companies, are exposed but analysis shows that SPVs are spreading really significant risk to many more investors.
It's an interesting counter to the efficient market hypothesis. "Everybody" knows about this debt. It's in the most public news outlets there are, and the word has been getting around. It's about as secret as Taylor Swift's concert schedules. Any serious investor knows about this debt.
And yet... the companies do this because it works. If they held this debt on balance sheet, the sensible assumption is that their stock values would take a rather substantial hit, and they could face other sorts of scrutiny. It works like pull-in sales works. It works like channel stuffing works. It works even when everybody knows that's what's happening. It works even when everybody knows that everybody knows that's what's happening.
There's something broken here. In an era where AIs move millions upon millions of dollars around because of some blip of a headline somewhere and every AI improvement of any kind is immediately scrutinized for its ability to be used by the financial system, it is completely incredible that the system doesn't know and react to these things. I'm not sure what's broken. My first best guess would be the increasingly mindless investment via index funds in pensions and the slow-but-ever-increasing ability of financial engineering to abuse that mindless investment, but I call that a "guess" for a reason. Possibly there's still a lot of really stupid AIs hooked up to the stock market that just look at the most basic of numbers and are easily fooled by this? But who is running such a precise combination of "huge" and "stupid" on the market? I dunno. Something's weird here.
Something doesn't quite smell
right about this story. Here's a key paragraph from the Nikkei story that this Futurism story re-tells:
> Companies disclose such future debt not in their balance sheets, but in annotations to their quarterly financial statements. This is a legitimate practice under accounting rules, but may make it difficult for retail investors to recognize risks.
Does that justify a "tries to hide" headline?
This is also one of those cases where the headline is free but the details are behind a paywall.
I do think the story itself is notable, but I expect the discussion is going to lack some nuance.
Really feels like the govt + industry, through protectionism and fear-mongering, are propping up a "Too big to fail" situation.
Long term, I think the best thing the economy could do is to make training on model outputs fair-use, as suggested by Ben Thompson[1]. Short of that, the companies should enter into distillation agreements with other US labs to let them make near-Fable models.
As it stands now, the companies want to hold all the upside. While also being culturally so safety focused - "only we have the right to regulate this" that its IMO counterproductive to US leadership in AI.
A different universe where X.ai, Meta, and everyone were also building Fable competitive open weights models - because they can distill - would probably be better for the US long term. But there's too much capital on the line right now behind OpenAI / Anthropic for them to do this.
Wouldn't improving LLM efficiency make them even more useful across the board, then they can enjoy the nice economies of scale?
The plan is to have LLM working completely autonomously, in that case, the more resources you have, the better.
Perhaps people will use local LLM to ask questions, or coders use them for their personal projects, but that's not where the real money is.
We haven't even started with a lot of things were we need a lot more compute:
Your real personal agent which knows you and helps you like "good morning elmer2, your calendar invite for dinner is today, you will need to leave at 18:18 if you want to use your normal public transport route per train. I put an alarm in your phone for you"
Agents to agents
Agentic teams.
Finetuned models for everything like Java/spanish coding model.
Very long term research like multiply hours or days or weeks and plenty of these in parallel.
Given how heavily subsidized it is at the moment, the efficiency isn’t as important. Typically efficiency would give you more at lower cost, but with token prices so removed from actual cost that plays less of a role here.
If the inference gets an order of magnitude cheaper, labs can afford to subsidise an order of magnitude more usage for the same marketing cost. So that part of usage will, if not accelerate with efficiency, at least still grow linearly with it. And there is a substantial amount of usage at or above true costs - everyone using a 3P harness, everyone on enterprise contracts, and everyone self-hosting an open weights model in a 3P cloud.
I assume the reference was to the Jevons paradox, as described in Jevons' 1865 book, "The Coal Question". Watt's steam engine massively increased the efficiency of coal in steam engines, which increased the use of coal fired steam engines, which increased coal consumption.
The (any!) comparrison to photovoltaics is
not acurate.Photovoltaics (PV) are primary energy producing infrastructure that produces its own fuel and is now verticly integrated into it's own supply chain, nothing other than life itself posseses this atribute.
AI, is exceptionaly likely to work in exactly the opposite fashion and take its host out as it goes down.
PV had to prove that it was truely indispensable and also prove to have realistic prospects for improvement and volume production before investments were made, and then prices came down.
AI has proven that it burns more money faster than anything else, ever.
I will admit that I am an early adopter of solar, but an AI refusenic, but still there is no reasonable comparison of AI and
anything outside of religion.
I disagree in a way, part of the reason they can't really succeed at the moment is because it's way too expensive to really deploy at scale for most companies, but even for those AI companies themselves. If they can make business access subsidized/cheap the same way pro/plus/max/whatever plan are for regular users while still being profitable, this can work out. The other solution is if they do reach that "it's so super smart it's reinventing the world every day", but that one is much more of a maybe possibly one day.
What they can't do is the rug pull of pricing like Fable did, hoping for profitability while playing the "it's so super smart" card. It's very profitable, but customer will be very happy to leave for cheaper pasture and that's why the recent news about this or that cheaper chinese models make headlines.
Essentially, the rush now is "if I make it a boring profitable company I'm not worth a trillion AND i'm overshadowed that plays the singularity card even if they're bullshitting"
You do realize "subsidizing" means charging less for something than it costs to provide, right? So they'll lose a dollar on every sale, but they'll make up for it in volume? E2E is usually where the profit comes from. If they're subsidizing getting regular users on board (pro/plus/max), and they're subsidizing to get businesses on board (massive deploys), where can the profit possibly come from without a pricing rug pull?
I do, my point was answering to the "if it comes so cheap that" they would stop losing money of that, they would still need to subsidize for acquisition or some big clients or for rush times. It's the all-you-can-eat-buffet strategy.
I'm not saying I see them going that way or that I would, but at least THAT would possibly work.
Thank you for the clarification. I figured I was missing something in your meaning. Although "if it comes so cheap that..." means it will probably come that cheap for everyone, and the margins wouldn't be there in the end because a price increase to take advantage of those big clients / rush times might lose the clients. I think we agree that their chances of becoming profitable don't look great.
Jevon's Paradox ("As efficiency of resource use increases, usage of the resource increases") says otherwise. One things become more efficient, we can use them in lots of ways that would not have been viable before, driving up usage.
The crux of the problem is models are not evolving anymore, they're iterating. Cheaper, faster, better. We're only seeing better, and that's because there is fierce competition with massive debt behind it. The "cheaper and faster" part is all taking place with local models. All of this adds up to a big red flag for a bubble.
If they continue investing in compute, memory, memory bandwidth, network infrastructure, etc. it makes a relevant contribution of progress in all of these fields which I will leverage.
A small form factor PC with 100gb fast memory and being able to run something like sonnet or opus level LLM would be massive.
I have so many things i want to do and still sitting it out due to cost.
Lots of people didn't invest in mortgage backed securities but still got screwed in 2008. When something is systemic, you don't have to be directly exposed to be effected when it goes sideways.
Is this in practice what's going to happen, or are (1) the prices going to hike (and never go down) for the consumer and (2) the memory companies will just continue doing what they already do because they're still selling their old shovels to the gold diggers?
I don't see how this will benefit the consumer, but I might be missing some second order effect?
I read somewhere that the memory companies were massivly pushed for lowest prices especially by companies like apple.
I want to hope that this money will lead to more capacity, more R&D and lower prices in the long term again.
Nvidia would have changed its GPU strategy a long time ago if the demand wouldn't be real. They still can afford the GPU prices. But memory is not a monopoly.
For memory though i do assume a lot more people and companies want a massive amount more memory than ever before. I have 64gb in my pc for a few years now, i was quite happy with that. It became a no brainer. But today? Hey give me 100, 300 and even more. I really want to run bigger LLM models locally.
If hyperscalers flop, and there’s a good chance they will, memory and disk prices will crater. They are historically the most volatile asset in tech. If Samsung, Micron et al can’t sell to hyperscalers they will switch back to consumer, because they can’t just turn off a memory fab without losing billions.
only that is not so. disk prices maybe, the memory will not be available to consumers because it's a tech that makes sense only for datacenters full of GPUs and massive power/cooling. by now all fabs have converted to it, there might be a lot of HBM capacity freed but no consumer devices that can use it. retooling all fabs to produce consumer level memory will take a lot of time if they even do it at all instead of just pushing for consumers to rent datacenter capacity directly.
all three major remaining players in the mem sector have already tried in the past to collude for memory price fixing.
this is the first time I'm rooting for chinese chip tech to reach more or less parity.
That switch will actually take a few months, technically speaking. What may delay it is manufacturers strategizing to avoid oversupply and trying to minimize their loses from all the investment in tooling they cannot longer repurpose. It will be a logistical and financial nightmare for them too.
Yes that is unfortunate for sure don't get me wrong this affects me but the overall benefit will still be bigger i assume.
10 years ago i watched a talk about the problem of compute vs. memory. Compute increased significantly while memory speed did not.
This gigantic investment will solve this problem.
So either this blows and we will have way too much capacity which will lead to cheap and mass amount of memory for everyone + cheap GPUs again OR AGI. So win - win.
Exactly, this is how capitalism works. Let them shoot for the moon and let them fail. Worst case their over-valued assets are liquidated and continued on with at a more reasonable valuation. Just make sure they play by the rules and don't make new rules in the name of national security, ie boxing out open source.
You could say this but the previous administration was not talking about taking a 10% share in both companies; this one is (thanks to Sam Altman's very personal lobbying of the president)
Except this is not how American capitalism works at this scale, and it’s ridiculous to think their debt isn’t your debt when you have the entire country’s history to look back on and count the numerous government bailouts.
There are several ways that ordinary investors and even simple pension holders could end up stuck with the downside of this.
The debt risk hidden in CDOs wasn't your debt either but if you had a pension plan, the crisis absolutely cost you money you would have earned, and in many cases pension fund values dropped by five to ten per cent within a year.
The SPV/CDO comparison being made is by no means exact, but hidden debt at this scale surprising analysts tends to cause problems. If more institutions are severely exposed than anyone thought, it is bad.
Especially since any success strategy is predicated on literally unbelievably rosy predictions.
AFAIK they have heavily relaxed the rules for IPO. Pension funds are practically forced to buy from the top-100 companies, and these companies risk crashing much more than the others.
SpaceX value is already lower than at launch. If this costs are externalized to the common public, this will be your debt.
All these companies are too big to fail, in an environment where you can buy pardons and laws.
Hell, a 3T$ crash will have global repercussion and probably partially crash many other countries, too.
Luckily so far only one index changed its rules to cover SpaceX and what the AI IPOs would need.
I don't know how specifically significant SpaceX being lower than at launch is, because actually most IPOs underperform the market and their own targets for the first three to five years. What is happening to it is not that unusual; its overvaluation is.
I do think there is a major risk here, and ordinary investors and pension holders will be hurt.
I am not sure any individual AI company is too big to fail, though probably one of the big two will be rescued, most likely Anthropic. I think OpenAI will fail, and it'll be stripped for parts. As will Oracle, who are overexposed to it.
If it crashes the global economy will crash and they will have to print money for a bail out which means another 30% increase to the price of everything
We also may be at the wealth inequality level where prices become… weird.
If there is really only a few dozen people doing the buying and the selling at the top on a weighted basis, then the prices are whatever they convince themselves of.
As long as this debt does not make it into life insurance and pension funds, we are fine. The trouble is that private credit is taking control of some life insurance companies and off-loads this debt to these. When these fail, it will become everyone's problem.
> Risks to financial stability may also stem from entities with particularly high exposure to private credit markets, such as insurers influenced by private equity firms and certain groups of pension funds. The assets of private‐equity‐controlled insurers have grown significantly in recent years, with these entities owning significantly more exposure to less‐liquid investments than other insurers
https://www.imf.org/-/media/files/publications/gfsr/2024/apr...
https://www.imf.org/-/media/files/publications/gfsr/2024/apr...
Couldn't it be a problem given the concentration of the S&P in these companies?
At this point these companies make up a huge portion of 401k's for a huge chunk of Americans. How would it affect retirees if they dropped 40-50%, likely taking the market with them?
I suggest looking into “EQL”, or better yet, just replicating its index by taking a position in the 11 XL* sector funds from SPDR, allocating equal weighting to each. One will end up with one’s equities equal weighted by sector and with plenty of large cap exposure, as opposed to the pronounced mid-cap tilt found in whole market equal-weight strategies.
Personally, I drop the financial sector entirely (Thomistic prohibitions on usury) which leaves an even 10 funds which is easy to allocate mentally and in practice. For example, assuming a 60/40 allocation where one is holding the lion’s share in equities and the remainder in bonds (I substitute with a combination of gold, crypto, cash, and Swiss Franc here), one would allocate as follows:
XLC 6% XLY 6% XLP 6% XLE 6% XLV 6% XLI 6% XLB 6% XLK 6% XLU 6% XLRE 6%
(Note that XLF is consciously not taken as a position here, decide if it’s right for you. The Mortgate REITs which would make XLRE problematic are in XLF per the sector selection rules)
The remaining 40% is bonded debt if you are fine with usury, or some sort of asset negatively or neutrally correlated to equities.
It’s an interesting thought. The growth is so extreme that if the S&P 500 fell 50% today it would reach levels last seen in 2022. Given that the timespan is so short, I’m honestly not sure it would be as bad for 401ks as people expect unless all of your investment was concentrated in the last 4 years.
I suppose it’s worse if your calculation is, “I’ll retire when my 401k hits $X absolute value,” but I think most people just retire at a certain age instead with risk spread across decades.
During the dot-com crisis. Nasdaq fell around 78% from its peak and S&P by around 49% so it isn't unprecedented (ironically has both aspects of being both tech and are within the same time-era)
It created an actual recession albeit thankfully short one for the case of dotcom (sadly not for 2007) and a really recessionary environment which causes unemployment and just straight up fear and panic.
I do understand what you are talking about and overall in long term, perhaps things flatten out but atleast speaking financially so, its better to be on a smooth sailing road rather than insane ups and downs with retirement money if preferable.
> I think most people just retire at a certain age instead with risk spread across decades.
The issue in my opinion is with people near that certain age you mention and who retire in the time during boom just before bust. They would then get the 50% hit on their savings instantly with an recession/inflation/unemployment environment which in my opinion might be genuinely devastating.
(supposing that they had their investments in stocks, I wouldn't consider that any retiree would have all their money in stocks but there have been some other comments which show a sizable amount, @kipchak's comment shows 50% stock for retirement. so a 50% shock on top of that could lead to a wipe out of 25% of your retirement fund which is honestly still pretty crazy.)
> (supposing that they had their investments in stocks, I wouldn't consider that any retiree would have all their money in stocks but there have been some other comments which show a sizable amount, @kipchak's comment shows 50% stock for retirement. so a 50% shock on top of that could lead to a wipe out of 25% of your retirement fund which is honestly still pretty crazy.)
What do you think the cost would be for protective puts per $1M of equity exposure of a portfolio on a monthly basis? "Bubble Insurance" if you will. Going 100% bonds is simply intolerable for the time window for most retirees, but an equity wipeout of this magnitude is equally intolerable.
NVidia makes up 7.5% of the SP500. If it lost 50%, it would be a 3% loss for the index. The concentration is bad, but it would not cause a drop of 50% retirement funds by itself. If you take an all world index, it's even less.
Still, if NVidia lost 50% of their market share, we would probably see a big collapse of the stock market.
EDIT: to note, the top ten companies in SP500 make up an unprecedented concentration but they're not "mostly AI".
It is unlikely that a 50% drop in NVidia wouldn't be paired with a significant drop in the valuation of every other company heavily invested in AI.
"This tree only makes up 0.0001% of the forest. If it is lit on fire, the forest will be fine"
Regarding unprecedented concentration, wasn't the nifty fifty era comparable for the top 10, about 40%?
Idunno man. Am I the only one that remembers the day the first DeepSeek model came out?
It wasn't like, "Nvidia took a hit and everyone else was fine". It was more like, "One or two companies were fine, and ALL others took a hit"
how are they not “mostly AI”?
I don’t think concentration risk is itself overly concerning. The nature of a market cap weighted index means it will always be heavy on whatever is currently trending. You’ll certainly be hurting if your plan is to retire at the top of the market with just enough, as the inevitable downturn will hammer your portfolio down into not enough. So invest until you have enough to handle volatility or a lost decade with a dip and slow recovery.
Even if theres a massive drawdown it will recover in the medium term (and in the short term is a great buying opportunity).
For the people who are close/early to retirement and can't do that, well, they need to manage sequence of returns risk.
Edit: I think some ppl might interpret this as me being bullish on the SP500. I'm not, I'm bullish on everything evens out and returns to the mean.
retirees arent suppose to have their active retirement funds in stocks dude. Any financial advisor with a brain would not make such a ridiculous asset allocation error.
Regardless of whether it's a good idea, it absolutely happens and as a result would impact retirees, both in individually managed accounts and target date funds. For example here 70+ are 45% equity.[1] TROW retirement 2020 funds are about 50% stock, for example, and only decrease to a floor of 30%.
https://workplace.vanguard.com/content/dam/inst/iig-transfor... (page 78)
For those who stick to a meaningful asset allocation (e.g. 60/40, 80/20, etc), this does not pose significant problem -- they would not be buying much stock in the last 3 years. Instead, they would be buying mostly fixed-income. Probably mostly in 401k/IRA accounts.
But if their debt goes bad, isn’t that debt the very bonds that make up the other part of those asset allocations?
Typical total bond market fund like BND is ~70% in USG -- pretty solid:
https://investor.vanguard.com/investment-products/etfs/profi...
Retirees relying on short term equity returns to cover expenses only have themselves to blame.
I'm not familiar with 401k rules but presumably they get a choice of markets and products?
If one is over concentrated its easily avoided.
The problem some have pointed out is that these companies are such a huge portion of the market right now.
The sound advice for the past decades has been, just invest in a low-cost ETF tracking the S&P instead of picking stocks to minimize risk and invest in the market broadly.
So a huge number of people have done that, believing they're diversified, while tech makes up 40% of the index.
Yes you could sell your S&P and find things to invest in least likely to be impacted by a potential bubble, but your average 9-5'er with automated contributions to their 401k is probably not sophisticated enough to do that.
And that's assuming only these companies would be affected if there was a massive draw-down in tech/AI related stocks. We haven't really seen a situation like this before, so it's not easy to predict what effects there might be in the broader economy.
no one that needs to rely on their investments for their actual retirement still has them in equities. theres a reason target date funds automatically adjust asset allocation as it nears its target date. you should be in majority bonds and cds well before your actual retirement date.
The employer selects a financial company to manage the 401k. When you switch jobs, you can roll the 401k from the previous employer into the new one, or into an IRA (Individual Retirement Account).
Usually the financial services company will offer several options: more aggressive/high risk, or less aggressive/lower risk. Most people will just go with whatever is the default option.
So much of the American S&P 500 is dominated by handful of companies that the risk is not that easy to avoid. If or when the AI bubble pops, it's going to take down a lot of the economy with it. You can direct your retirement savings into the lowest yield/lowest risk assets offered by the firm, but you'll forego whatever growth happens in the mean time.
Will the bubble pop next week? Next month? Next year? Who knows. Timing the market is incredibly difficult.
There's a famous quote, attributed (perhaps apocryphally) to John Maynard Keynes: "The market can remain irrational longer than you can remain solvent."
The whole idea of a pension fund is that you don't need to time the system it is the system.
Like my country pension scheme. It went through ups and downs for a hundred years but has always come on top. All you need is a long horizon and a trillion dollars and you basically can't lose.
There is ZERO chance the modern oligo-kleptocracy isn't going to socialize the losses onto the little guy
> When these fail, it will become everyone's problem.
Debt is senior to equity. For private credit to start taking haircuts, the equity has to have already gone to zero. At that point, this will already have been everyone's problem for some time.
Are you suggesting that holding private credit assets is relatively risk free?
Equity is a risky asset, so equity being wiped out should not be a surprise to anyone, but life insurance and pension funds failures is indeed a public problem.
There's a wide spectrum of 'private credit', and it varies from low-risk to quite high risk, depending on the debtor. In the case of "AI Companies", their debt is low-risk, but many are creating special-purpose-entities which will build and own some or all of their newer datacenters. Those datacenter companies are issuing a great deal of somewhat risky debt; with the exact level of risk depending on the off-take agreement.
I disagree - high leverage inherently makes systems less stable.
You probably meant to say that practically, high leverage tends to leak into companies of public interest. For example, when high net worth individuals start trimming their private credit holdings, which eventually end up with insurers. That is why the regulators have to watch carefully that it does not happen.
> As long as this debt does not make it into life insurance and pension funds, we are fine.
I think for small to even large numbers you are correct, but given how yuge this debt amount is a broad-based default will probably cause a contagion. I will not predict how far and wide.
You say this as though every company doesn't take on debt, and all debt isn't a risk. I'm sure you have some much riskier debt than ChatGPT already in your portfolio, and interest rates are adjusted by relative as judged by the market. I'm sure some debt will fail, but certainly all of it won't, and while anything could cause a market crash saying "when these fail" holds a lot of incorrect assumptions.
bingo - if the firms holding the debt keep holding the debt & the debt doesn't get passed to other entities - the system will be fine.
if say meta owes 720Bn, they wouldn't have trouble paying that back in 10 years.
this doesn't take away the fact that 'a.i' right now is a bubble.
Do they? Is a company with $200 billion annual revenue and earnings (EBITDA) of $100 billion having $420 billion of off-balance-sheet debt really staggering?
In many other industries that would be a perfectly normal amount of debt to have. It's only unusual because we are used to tech companies having so much cash on hand they don't know where to put it
These companies have valuations reflecting a debt light business. At a minimum, 420 billion in debt is enough to change the stock price by 10-20%. If the company plans to add another 400 billion in debt you need to give it the side eye.
If 50 billion in revenue is from other companies debt spending… then You have a problem.
> These companies have valuations reflecting a debt light business.
Sorry, but this doesn’t make sense. The valuations of these companies reflect their growth.
In finance there’s nothing inherently virtuous about a “debt-light business”. It’s all an allocation decision based on how you expect to grow relative the cost of that growth.
Try and reframe it: are cash-heavy businesses given a premium?
The pertinent comparison in valuations is debt vs equity, not debt vs cash as you noted.
My point was more of an exercise to point out that finance is about mutating resources. A lot of cash can be a good thing or a bad thing. Same for debt. There’s nothing inherently bad about levels.
> they don't know where to put it
ohh, their accountants just dont know where debt goes on the balance sheet. thanks for clearing it up
It is not just that they have the debt, it. is they are trying to hide the debt. Why would a legitimate company try to hide their debt?
There are many reasons to use subsidiaries for things like this, like to invite outside investment, ringfence risk, cede operational risk, and many more.
This is all like CFO 101 type stuff, and not nefarious. I find it amusing that people assume the worst for things they understand little about, rather than trying to learn.
Maybe the best way I can explain it to the programming crowd is this: imagine how ridiculous it would sound if outsiders were saying that Google was on the verge of collapse because its codebase has billions of lines of code.
> There are many reasons to use subsidiaries for things like this, like to invite outside investment, ringfence risk, cede operational risk, and many more.
Agreed, there are many valid reasons to have subsidiaries of course.
The issue is rather with the fact that we are having the assumption that the threat is outside rather than inside and so systems with mechanisms to be less transparent are far more prone to this risk.
It has been academically shown that most corporate/ white paper scams aren't done from outside but rather from inside the company itself through genuine structures and incentives which go wild. (Something shockingly visible in AI space), Enron's example also comes to my mind.
The best way I can explain it to the programming crowd is this: Imagine how ridiculous it would sound if you are ranked with how many lines of code you ship and how much token you would spend and so we end up with tokenmaxxing and hearing stories about people literally burning tokens in innovative ways because they want to get on top of a leaderboard. Oh wait, it is already happening or has happened.
Generally speaking, It is preferable to be transparent with debt and other things rather than not especially so for long term because sooner rather than later you might get caught. Obviously if there is some stuff which prefers from ringfencing risk then sure.
Also as I spoke of Enron, but the exact structure was used by Enron as well as @fzeroracer discusses in their comment[0] so it might be a genuine question.
[0]: https://news.ycombinator.com/item?id=49023596
"There are many reasons to use subsidiaries for things like this, like to invite outside investment, ringfence risk, cede operational risk,..."
In other words, they are intentionally deceiving investors and hiding the risk from them. That does not sound like CFO 101, it sounds like fraud. But if grift is your business, I guess those are as valid reasons as any.
She's a witch!
> is they are trying to hide the debt.
They aren't hiding it though. The contracts are recorded in regular filings.
Rope a doped with cope. Maybe you should buy some $ORCL?
https://asia.nikkei.com/business/technology/five-us-tech-gia...
"Companies disclose such future debt not in their balance sheets, but in annotations to their quarterly financial statements. This is a legitimate practice under accounting rules, but may make it difficult for retail investors to recognize risks."
"Today's AI industry is partly supported by demand generated by circular investment. Nvidia and tech giants invest in data center operators and AI companies, with that money then turning into GPU and cloud usage fees. Actual demand is difficult to see, increasing the likelihood of over investment in data centers."
Retail investors shouldn’t be investing in individual stocks outside of industries they understand well. Following GAAP is the definition of not hiding the obligations.
Enron also didn't violate GAAP so by your definition, no financial wrongdoings...
Maybe GAAP should be changed? Professional investors have the time to comb through the small print. These off balance sheet activities seem designed to bamboozle the small investor.
>but may make it difficult for retail investors to recognize risks
Ok, so just to be clear: institutional investors are (a) the ones investing the large proportion of capital in these companies and (b) are well equipped to decipher financial statements. The idea that any significant amount of retail investors have even seen a financial statement, let alone is making decisions based on their analysis of a financial statement, is laughable. And, even then, if someone is putting in that effort, then presumably they're not going to get tripped up by a legitimate practice that they ought to specifically be looking for given the context.
And all of that doesn't even take into account that every article discussing the financials of AI firms over the last half a decade have been pointing out these dynamics. We're literally in a thread discussing this exact dynamic. Retail investors are certainly far more likely to make investing decisions based on these kinds of articles and threads than they are based solely on independent financial statement analysis that they're conducting. At a minimum, I think anybody taking any of this seriously has gotten the hint by now.
If it comes out that these firms are committing straight-up fraud, then there will be a lot more to discuss. But, as of now, the sentiment is that these firms are behaving perfectly legitimately, just abnormally and maybe irresponsibly compared to their historical context. If an investor isn't equipped to handle this kind of analysis under these circumstances, then I'm not going to feel too bad if they lose their money "investing" when they're really just gambling.
Because they have even more debt than the debt we assume they’re trying to hide
they aren't trying to hide anything, those are accounting rules that are applied to the letter. I'm feeling like I'm taking crazy pills whenever I see this stuff about AI, your hate boner for a specific technology shouldn't trigger you saying things that are provably untrue.
Are they really "trying to hide" this debt? I think it's pretty common knowledge that a lot of these companies are using debt/bonds for funding. The debt not showing up where the author wants is a reporting formality not an attempt to hide it.
I think the point is that it’s not showing up on the standard financial filings. If you were to pull the annual reports for these companies, you wouldn’t see it. That doesn’t mean it’s impossible to find it. Obviously, it is otherwise the article wouldn’t have been written. But you’re going to have to go the extra mile. To be clear, none of this is illegal. It’s just covered in the advanced CFO accounting class.
It's not hidden at all. Financial blogs very accessible to laymen like Matt Levine's Money Stuff have talked about this structure months ago. If you are an investor and surprised by this news you weren't sufficiently prepared and shouldn't have been investing in the first place.
What's the purpose of keeping it off the balance sheet if not to hide it?
Take-or-pay contracts appear as "contractual commitments" in 10-K. They are not hidden. That's the way they are reported in all industries where take-or-pay contracts exist. There's nothing nefarious about it.
If it didn't matter, why would they bother jumping through hoops to keep the debt off their balance sheet?
In the run-up to 2008 a big factor in the bubble forming was that poor quality loans were packaged in a way to hide the risk in those investments. I'm not expert enough in finance to know if it's the case now, but we do know that clever accounting to hide debt can lead to the incorrect valuation of assets, potentially leading to financial ruin.
They're not jumping through any hoops, I think they're simply complying with reporting requirements. It's not on their balance sheet because being recorded as strait debt would itself be misleading. My understanding is that these sort of off-balance sheet "debt" is mostly in the form of deal terms that may or may not be expressed at some point in the future.
An analogy that comes to mind is when companies used to book future sales in the present. They got in trouble for this and is now forbidden. I recall reading that one deal had terms that transferred assets if certain conditions were not met. If terms-based debt should be booked now, then terms-based assets should as well. This stuff makes my head hurt.
Either way, as long as it's not hidden (and it's not for the public companies), then it's fine.
> The debt not showing up where the author wants is a reporting formality not an attempt to hide it.
Couldn't you characterize Enron that way? The liabilities are there, you "just" have to look at Raptor II or whatever!
You couldn't just characterize them that way, they are using literally the exact same setup that Enron did in order to hide the massive amount of debt and liabilities they had that eventually sunk the company. People jumping in to defend this as being totally legit or not-fraud seem quite insane to me: companies should not be trying to hide these things to pump up their stock prices or to entice investors that otherwise might not see it.
Yeah I think it went through the press on mass eh?
And even if you look at the debt, even companies like meta make 200 billion revenue in 2025 alone.
Isn't it good that these companies with these massive massive deep pockets invest?
One of Meta's SPVs building a data centre for them, Meta own only 20% of it; 80% is owned by other investors. That's the issue here; the market thinks that only these handful of money-go-round FAANGs/Mag7 companies, are exposed but analysis shows that SPVs are spreading really significant risk to many more investors.
It's an interesting counter to the efficient market hypothesis. "Everybody" knows about this debt. It's in the most public news outlets there are, and the word has been getting around. It's about as secret as Taylor Swift's concert schedules. Any serious investor knows about this debt.
And yet... the companies do this because it works. If they held this debt on balance sheet, the sensible assumption is that their stock values would take a rather substantial hit, and they could face other sorts of scrutiny. It works like pull-in sales works. It works like channel stuffing works. It works even when everybody knows that's what's happening. It works even when everybody knows that everybody knows that's what's happening.
There's something broken here. In an era where AIs move millions upon millions of dollars around because of some blip of a headline somewhere and every AI improvement of any kind is immediately scrutinized for its ability to be used by the financial system, it is completely incredible that the system doesn't know and react to these things. I'm not sure what's broken. My first best guess would be the increasingly mindless investment via index funds in pensions and the slow-but-ever-increasing ability of financial engineering to abuse that mindless investment, but I call that a "guess" for a reason. Possibly there's still a lot of really stupid AIs hooked up to the stock market that just look at the most basic of numbers and are easily fooled by this? But who is running such a precise combination of "huge" and "stupid" on the market? I dunno. Something's weird here.
Would have been nice if the article had any substantive facts in it
The article is a very shallow restatement of the conclusions in this paywalled piece: https://asia.nikkei.com/business/technology/five-us-tech-gia...
Yup, at least a table of the on-books and off-books debt of the top 5 AI-building companies would be nice
"No you guys it isn't actually an issue because it isn't."
Why?
"Because it isn't; okay?!"
oh ok.
“You found it didn’t you?, then we weren’t actually hiding it, so please stop looking into our finances too much”
Something doesn't quite smell right about this story. Here's a key paragraph from the Nikkei story that this Futurism story re-tells:
> Companies disclose such future debt not in their balance sheets, but in annotations to their quarterly financial statements. This is a legitimate practice under accounting rules, but may make it difficult for retail investors to recognize risks.
Does that justify a "tries to hide" headline?
This is also one of those cases where the headline is free but the details are behind a paywall.
I do think the story itself is notable, but I expect the discussion is going to lack some nuance.
It's cute they think 'retail investors' are reading balance sheets in the first place.
Gosh it would be ironic if they were hiding the nuances behind the firewall in a way that makes it difficult for retail investors to read.
Only Oracle is in any kind of danger from their debt load, though. I have not checked SpaceX situation.
Meta, Google, Amazon, .. they can take the hit and go on.
Really feels like the govt + industry, through protectionism and fear-mongering, are propping up a "Too big to fail" situation.
Long term, I think the best thing the economy could do is to make training on model outputs fair-use, as suggested by Ben Thompson[1]. Short of that, the companies should enter into distillation agreements with other US labs to let them make near-Fable models.
As it stands now, the companies want to hold all the upside. While also being culturally so safety focused - "only we have the right to regulate this" that its IMO counterproductive to US leadership in AI.
A different universe where X.ai, Meta, and everyone were also building Fable competitive open weights models - because they can distill - would probably be better for the US long term. But there's too much capital on the line right now behind OpenAI / Anthropic for them to do this.
They're really in a bind IMO.
1 - http://stratechery.com/2026/whos-afraid-of-chinese-models/
> Long term, I think the best thing the economy could do is to make training on model outputs fair-use
AI outputs have been ruled as not even copyrightable, isn't that even better than fair use?
Probably - The issue is more about terms-of-service and whether any company wants to go to bat on a years-long legal battle over this issue
It won't pay off if LLMs efficiency gets good enough to make those data centers obsolete.
It's a huge gamble.
Wouldn't improving LLM efficiency make them even more useful across the board, then they can enjoy the nice economies of scale?
The plan is to have LLM working completely autonomously, in that case, the more resources you have, the better. Perhaps people will use local LLM to ask questions, or coders use them for their personal projects, but that's not where the real money is.
We haven't even started with a lot of things were we need a lot more compute:
Your real personal agent which knows you and helps you like "good morning elmer2, your calendar invite for dinner is today, you will need to leave at 18:18 if you want to use your normal public transport route per train. I put an alarm in your phone for you"
Agents to agents
Agentic teams.
Finetuned models for everything like Java/spanish coding model.
Very long term research like multiply hours or days or weeks and plenty of these in parallel.
This is going to sound a bit facetious, but I'm almost certain the Gmail / maps integration on my Samsung galaxy in 2014 did that.
Or: increasing resource efficiency may encourage even more usage, as happened with coal, oil and photovoltaics.
https://en.wikipedia.org/wiki/Jevons_paradox
Given how heavily subsidized it is at the moment, the efficiency isn’t as important. Typically efficiency would give you more at lower cost, but with token prices so removed from actual cost that plays less of a role here.
If the inference gets an order of magnitude cheaper, labs can afford to subsidise an order of magnitude more usage for the same marketing cost. So that part of usage will, if not accelerate with efficiency, at least still grow linearly with it. And there is a substantial amount of usage at or above true costs - everyone using a 3P harness, everyone on enterprise contracts, and everyone self-hosting an open weights model in a 3P cloud.
They improved the efficiency of coal?
Absolutely. Early engines were so inefficient that they could basically only be used in coal mines.
Massively. Extracting more useful energy from coal has always been a goal.
Very much unlike with software. Where the goal for long while is to burn as many resources as possible on end user devices.
I assume the reference was to the Jevons paradox, as described in Jevons' 1865 book, "The Coal Question". Watt's steam engine massively increased the efficiency of coal in steam engines, which increased the use of coal fired steam engines, which increased coal consumption.
Of coal use, yes. That's what inspired this: https://en.wikipedia.org/wiki/Jevons_paradox
The (any!) comparrison to photovoltaics is not acurate.Photovoltaics (PV) are primary energy producing infrastructure that produces its own fuel and is now verticly integrated into it's own supply chain, nothing other than life itself posseses this atribute. AI, is exceptionaly likely to work in exactly the opposite fashion and take its host out as it goes down.
When PVs got cheaper, more of them were sold.
PV had to prove that it was truely indispensable and also prove to have realistic prospects for improvement and volume production before investments were made, and then prices came down. AI has proven that it burns more money faster than anything else, ever. I will admit that I am an early adopter of solar, but an AI refusenic, but still there is no reasonable comparison of AI and anything outside of religion.
I disagree in a way, part of the reason they can't really succeed at the moment is because it's way too expensive to really deploy at scale for most companies, but even for those AI companies themselves. If they can make business access subsidized/cheap the same way pro/plus/max/whatever plan are for regular users while still being profitable, this can work out. The other solution is if they do reach that "it's so super smart it's reinventing the world every day", but that one is much more of a maybe possibly one day.
What they can't do is the rug pull of pricing like Fable did, hoping for profitability while playing the "it's so super smart" card. It's very profitable, but customer will be very happy to leave for cheaper pasture and that's why the recent news about this or that cheaper chinese models make headlines.
Essentially, the rush now is "if I make it a boring profitable company I'm not worth a trillion AND i'm overshadowed that plays the singularity card even if they're bullshitting"
You do realize "subsidizing" means charging less for something than it costs to provide, right? So they'll lose a dollar on every sale, but they'll make up for it in volume? E2E is usually where the profit comes from. If they're subsidizing getting regular users on board (pro/plus/max), and they're subsidizing to get businesses on board (massive deploys), where can the profit possibly come from without a pricing rug pull?
I do, my point was answering to the "if it comes so cheap that" they would stop losing money of that, they would still need to subsidize for acquisition or some big clients or for rush times. It's the all-you-can-eat-buffet strategy.
I'm not saying I see them going that way or that I would, but at least THAT would possibly work.
Thank you for the clarification. I figured I was missing something in your meaning. Although "if it comes so cheap that..." means it will probably come that cheap for everyone, and the margins wouldn't be there in the end because a price increase to take advantage of those big clients / rush times might lose the clients. I think we agree that their chances of becoming profitable don't look great.
Don’t worry, they’ll get bailed out
Jevon's Paradox ("As efficiency of resource use increases, usage of the resource increases") says otherwise. One things become more efficient, we can use them in lots of ways that would not have been viable before, driving up usage.
The crux of the problem is models are not evolving anymore, they're iterating. Cheaper, faster, better. We're only seeing better, and that's because there is fierce competition with massive debt behind it. The "cheaper and faster" part is all taking place with local models. All of this adds up to a big red flag for a bubble.
I guess "try to hide" means to be posted about all over the news weekly.
Alternative title: Memory, GPUs, and SBCs are about to become affordable again :)
And? Its not my debt.
If they continue investing in compute, memory, memory bandwidth, network infrastructure, etc. it makes a relevant contribution of progress in all of these fields which I will leverage.
A small form factor PC with 100gb fast memory and being able to run something like sonnet or opus level LLM would be massive.
I have so many things i want to do and still sitting it out due to cost.
Lots of people didn't invest in mortgage backed securities but still got screwed in 2008. When something is systemic, you don't have to be directly exposed to be effected when it goes sideways.
Is this in practice what's going to happen, or are (1) the prices going to hike (and never go down) for the consumer and (2) the memory companies will just continue doing what they already do because they're still selling their old shovels to the gold diggers?
I don't see how this will benefit the consumer, but I might be missing some second order effect?
I read somewhere that the memory companies were massivly pushed for lowest prices especially by companies like apple.
I want to hope that this money will lead to more capacity, more R&D and lower prices in the long term again.
Nvidia would have changed its GPU strategy a long time ago if the demand wouldn't be real. They still can afford the GPU prices. But memory is not a monopoly.
For memory though i do assume a lot more people and companies want a massive amount more memory than ever before. I have 64gb in my pc for a few years now, i was quite happy with that. It became a no brainer. But today? Hey give me 100, 300 and even more. I really want to run bigger LLM models locally.
The thing is they actually pushed the price tags of memory and disks high so we wouldn’t be able to afford it. Unless ofc you rent from them.
If hyperscalers flop, and there’s a good chance they will, memory and disk prices will crater. They are historically the most volatile asset in tech. If Samsung, Micron et al can’t sell to hyperscalers they will switch back to consumer, because they can’t just turn off a memory fab without losing billions.
only that is not so. disk prices maybe, the memory will not be available to consumers because it's a tech that makes sense only for datacenters full of GPUs and massive power/cooling. by now all fabs have converted to it, there might be a lot of HBM capacity freed but no consumer devices that can use it. retooling all fabs to produce consumer level memory will take a lot of time if they even do it at all instead of just pushing for consumers to rent datacenter capacity directly.
all three major remaining players in the mem sector have already tried in the past to collude for memory price fixing.
this is the first time I'm rooting for chinese chip tech to reach more or less parity.
That switch will actually take a few months, technically speaking. What may delay it is manufacturers strategizing to avoid oversupply and trying to minimize their loses from all the investment in tooling they cannot longer repurpose. It will be a logistical and financial nightmare for them too.
Yes that is unfortunate for sure don't get me wrong this affects me but the overall benefit will still be bigger i assume.
10 years ago i watched a talk about the problem of compute vs. memory. Compute increased significantly while memory speed did not.
This gigantic investment will solve this problem.
So either this blows and we will have way too much capacity which will lead to cheap and mass amount of memory for everyone + cheap GPUs again OR AGI. So win - win.
Exactly, this is how capitalism works. Let them shoot for the moon and let them fail. Worst case their over-valued assets are liquidated and continued on with at a more reasonable valuation. Just make sure they play by the rules and don't make new rules in the name of national security, ie boxing out open source.
Here in the US "let them fail" only happens to businesses without political pull, which I'm pretty sure these companies have.
Do you think the businessmen who sat on the dais at Trump's inauguration plan to just fail without getting him to put his small thumbs on the scales?
It has nothing to do with the administration. These companies did the same thing with the last, and they will do the same thing with the next.
You could say this but the previous administration was not talking about taking a 10% share in both companies; this one is (thanks to Sam Altman's very personal lobbying of the president)
It manifests in many different forms.
Except this is not how American capitalism works at this scale, and it’s ridiculous to think their debt isn’t your debt when you have the entire country’s history to look back on and count the numerous government bailouts.
> And? Its not my debt.
For the moment.
There are several ways that ordinary investors and even simple pension holders could end up stuck with the downside of this.
The debt risk hidden in CDOs wasn't your debt either but if you had a pension plan, the crisis absolutely cost you money you would have earned, and in many cases pension fund values dropped by five to ten per cent within a year.
The SPV/CDO comparison being made is by no means exact, but hidden debt at this scale surprising analysts tends to cause problems. If more institutions are severely exposed than anyone thought, it is bad.
Especially since any success strategy is predicated on literally unbelievably rosy predictions.
> And? Its not my debt.
Your view seems very myopic.
AFAIK they have heavily relaxed the rules for IPO. Pension funds are practically forced to buy from the top-100 companies, and these companies risk crashing much more than the others.
SpaceX value is already lower than at launch. If this costs are externalized to the common public, this will be your debt.
All these companies are too big to fail, in an environment where you can buy pardons and laws.
Hell, a 3T$ crash will have global repercussion and probably partially crash many other countries, too.
Luckily so far only one index changed its rules to cover SpaceX and what the AI IPOs would need.
I don't know how specifically significant SpaceX being lower than at launch is, because actually most IPOs underperform the market and their own targets for the first three to five years. What is happening to it is not that unusual; its overvaluation is.
I do think there is a major risk here, and ordinary investors and pension holders will be hurt.
I am not sure any individual AI company is too big to fail, though probably one of the big two will be rescued, most likely Anthropic. I think OpenAI will fail, and it'll be stripped for parts. As will Oracle, who are overexposed to it.
Thats a Elon Musk / Space-X issue thought not a Google and co issue.
How much real impact is this really though?
If it crashes the global economy will crash and they will have to print money for a bail out which means another 30% increase to the price of everything
But minimal real impact
Is it your contention, then, that the 2008 crash had minimal real world impacts?
Is the bubble bursting? Amount of the news about bad shape of companies highly invested in AI in the past days are quiet alarming or is it just bias?
This media frenzy can go on forever tho.
A Bear Sterns moment would be more solid. Oracle might ge the first to collapse if things go south, so we might be fine until then(?).
It would be better if you all read the article this article was referring to:
https://asia.nikkei.com/business/technology/five-us-tech-gia...
[dupe] Discussion on source: https://news.ycombinator.com/item?id=48987863
Is it time to short AI companies?
Market can remain irrational longer than you can remain solvent...
Simply choosing not to get involved might be most reasonable action.
In mice! It's already factored in to the price.
We also may be at the wealth inequality level where prices become… weird.
If there is really only a few dozen people doing the buying and the selling at the top on a weighted basis, then the prices are whatever they convince themselves of.
There's only one mostly AI company you can short right now and everyone's already doing it.
What is that company?
Oracle.
I don't think the GGP meant that one. But yeah, that is one too.
SpaceX is already 3x as expensive to short as the next biggest megacap. Good luck everybody.
Ummm, whose that?
with US Government owning huge chunks now "too big to fail"
bailout incoming
will make subprime crash seem like child's play
sure you won't be able to ever afford a home but we'll have tons of cheap super-hardware barely used