US tech giants' hidden AI debt swells 90% to $3.2T in 6 months
asia.nikkei.com
asia.nikkei.com
This is a significant proportion of the market. Having this amount of money not understood is harmful.
Is this money not understood by the tech giants taking on the debt? I imagine they understand exactly what’s going on.
I think the way they see it is that it’s riskier for them to miss out on timing this investment.
The other question is “where else should they put their money?” Tech giants before the AI boom made a heck of a lot of cash that they had nothing they could do with it. Remember before AI when the existential crisis was stock buybacks? That was a bunch of companies who had so much profit that all they could think to do with it was buy their own stock back.
I am sure there will be a bubble pop, correction, general downturn, or something like that, but the question remains how much it’s actually going to matter. If AI blows up in Microsoft or Google’s face they’re just going to write it off and spend a decade paying zero income tax, or whatever other clever corporate accounting gets them out of that pickle.
If you or your retirement is a shareholder in one of these companies, you'd rather they return $100 billion to you with boring buybacks or dividends than put it in an exciting AI investment that returns $1-$10 billion. Even though it didn't end up completely worthless, it's still a bad investment.
> If AI blows up in Microsoft or Google’s face they’re just going to write it off and spend a decade paying zero income tax
As a shareholder, you'd also rather they return $100 billion to you than put it in a bad investment that returns $0 -- even if that bad investment lets them write off $20 billion in taxes over the next 10 years you'd still be $80 billion better off.
> how much it’s actually going to matter. If AI blows up in Microsoft or Google’s face
An AI winter shouldn't be an existential threat to these companies -- but "the share price is 10% - 20% lower than the counterfactual where they didn't make these questionable investments" still represents a lot of lost wealth for shareholders -- billions (or trillions for the industry as a whole). A lot of that wealth is the hands of ordinary people, often through retirement investments in index funds.
E.g., if Google doesn’t invest in AI then AI will kill their search engine business. A lot of us don’t love the Google search AI Overview but now they’re competing directly with ChatGPT, and they’re essentially not a search monopoly anymore.
One way to look at a bust AI investment is the destroyed value of the stock like you mentioned, but perhaps that’s somewhat hindsight-biased? Perhaps the reality for some companies is that if they sit on their hands and do nothing they lose 40% of their company value and if they spend a ton of money on AI they lose 20% of their company value. Of course we don’t know the outcome yet.
AI has already killed their search biz. No one even clicks on results anymore. How many ads are they clicking on these days?
If you’re a retireer yeah you just want a steady dividend (or buyback).
But i mean for a 401k that won’t be spent from for 20+ years you absolutely want business to be investing into themselves. Otherwise go buy gold.
Inside a 401k dividends get reinvested, on top of that companies value income go up with inflation simply because they’re getting a similar cut of larger numbers.
That’s why long term stocks are a vastly better investment.
As a shareholder I'd rather them take the $100bn and invest it in growing the business further and generating $200bn. That's the more likely outcome anyway.
Most shareholders in Time Warner, for example, regretted giving 50%+ of the company to AOL shareholders, even if everyone was correct that 'winning the Internet' was the key to growing the business.
Maybe they are counting on taxpayers to bail them out if they crater in a 2008-scale failure.
That was with a US government that was significantly less indebted than it is now, both in real terms and as a percentage of annual GDP. The crude birth rate is about 30% lower than it was in 2007. The generation that was coming of age at the time is the first in American history to have a lower quality of living than their parents.
IMHO, unless there is some massive geographic and societal divider that stands between the likes of Sam, Dario, et al. and the rest of the country at the time of any sort of collapse, they'd be far, far better off just taking the losses and not bothering American society again. It won't be pretty if they come asking for a handout.
I hate this way of thinking. The parents, at the same stage in life, didn’t have magical devices they could hold in their hands that would answer any question ever asked in human history, for example.
Progress matters. Just because the average 20 year old at that time could rent a 10% smaller apartment than their parents could, doesn’t mean they would necissarily trade what they have access to in 2007 for what their parents could access in the eighties.
People in 2007 are vastly better off than in 1984.
You'd also reasonably imagine that every major financial institution on the planet in combination with several sovereign governments would collectively have a handle on the market and yet CDOs nearly bricked the global economy in 2008. History suggests your faith here is likely misplaced.
[1]: https://www.youtube.com/watch?v=iJ84XMTugFs
It isn't to your original question but in terms of giving a cost to misunderstandings of finance one needs look no further then American politics.
401k growth is wrongly the face of the American Economy when it should be more tangible attributes like the understanding of debt at the macroeconomic level.
So if this is in fact a bubble, nobody can really act differently until the market stops treating AI spending as a winning strategy.
If you’re a Japanese company making spanners in 1988 and you’re not speculating in land while all your competitors are, your profits are going to look terrible in comparison and the board will probably try and fire the leadership and replace them with people who’ll speculate in land.
This is part of the reason bubbles can go on for so long - there’s often a point where a lot of people sort of know it’s a bubble and it will end in tears, but they’re still incentivised to participate
It’s not like we have any other industry doing well (or could be made to do very well with such an investment that would also pay off).
Maybe a few trillion could allow us to overtake China as the world’s leader in battery technology and manufacturing…
But it would take more time, effort, and a lot of logistics. The payoff from AI is much sooner…
Well yeah, that's the most obvious way this plays out
I'm just spitballing here, so bear with me, but maybe increasing the value returned to the people doing the actual work that created the profit in the first place.
Tangentially reminds me of the Costco Hot Dog story. [0] We know that companies rather increase prices for consumers to have better balance sheets, than changing product strategies, and for sure we know actual workers get the shortest end of the stick.
[0] - https://en.wikipedia.org/wiki/Costco_hot_dog#Cultural_impact
Dividends.
Fair. I think everyone who is financially astute knows what's going on. I think pretending "oh, we didn't expect this" is at best, faux naieve. But, I think underlying "understand" is "don't care" -I think there is a segment of the investment industry which only looks to their own profit, not the "build a bigger market" outcome. And so, wiping 4T of debt out, with some recovery in the ashes can be sold as a win for them, if they can be in the right segment of the market.
I don't think it's a net win overall. I don't think driving worldwide RAM prices high was a win. Mind you, I bet RAM manufacturers are quite happy.
Look back at the “dotcom” bubble. And the Savings and Loan crisis. And the Subprime mortgage crisis.
This is going to be 10x all of those put together, minimum.
There's a game-theoretic way of modelling this situation called, from memory, a modified Dutch auction. The way it works is that the winner pays the difference between their bid and the next highest bid, and so on all the way down. It models the way in which players are motivated to extend themselves almost infinitely in an attempt to win something unwinnable (what are the success criteria for "winning the AI race", in other words how do you know when you've "won"? And what do you get for "winning"?).
Many things change in the world but not human nature, we are doomed to repeat this forever
Wtf are the downvotes. It's a question.
We don't live in that world anymore. The bailouts changed the world forever. Since 2007 there is no such things as "with their own money" when it involves a big company risking bankruptcy. It's no longer private money.
Chancellor on the brink
These are signed contracts for future spending. However, we don't know what the terms are for those contracts and how easy it is to get out of them. For example, with Anthropic and Google's contracts with xAI's compute, they signed a long term contract, but they can get out of the contract with 3 months notice. That isn't a good indication of future spend. The same goes for all of these $3.2T, so these numbers are sus at best.
One thing the tech giants are worried about are being left behind on expanding their compute. So what do they do? Sign a contract to agree to purchase $XT in compute over the next X years. It’s like buying an option - it’s a right, not an obligation to purchase.
If I were an investor and spent $10,000 to buy options to purchase $1,000,000 in Google stock, it’s not “$1,000,000 in hidden debt”
I don’t think this is how your average reader of this would interpret debt.
When I and most people think of debt they think a negative equity account balance that you pay off to a bank or lender like a mortgage or credit card. Not a 24 month apt lease because no equity is being exchanged.
In part because of the reason OP said which was severability which isn’t something you can do with owed debt.
Graebers debt demonstrates that this is the most common way to think about it.
To my lay understanding, debt means a loan you have to pay back, whereas liabilities are everything you're on the hook for.
The difference is whether it is an executory contract.
If I sign a contract committing me to spend $10m in the future, but the other party is obligated to deliver that $10m of services, that is an executory contract and it is not considered a liability under IFRS because the commitment is contingent on the other party performing.
You're mixing this up with property leases, which create a debt because the tenant has obtained a Right of Use (ROU) asset, which is offset by the debt (lease payments). Leasing a property for ten years gives you both an asset and an offset liability, but only once the lease has commenced.
An obligation to pay that is contingent on the other party's performance of a contract is absolutely not a "debt".
Now, once that data center is built and turned over to the client, that may create an ROU that would generate both an asset (the ROU) and the offsetting liability. But if it's some future data center that will be built, it does not.
Even then, once the data center is built, an ROU may not be created: if the commitment is for a million GPU hours, with the seller having some freedom on how to deliver those GPU hours, that does not create an ROU. If it's for 100 specific GPUs for 10,000 hours each, it does, because it's about the right of use for those specific pieces of property.
For lease accounting, the asset and debt are only created at the point the lease begins, not the signing of the contract, so your reference to IFRS isn't relevant here.
This is very much like the 2008 disaster. You effectively have debt laundering happening whereby speculative income / debt is manufactured into higher grade than it is by passing through multiple parties / steps / debt-like vehicles that look independent but in fact incorporate cumulative risk built on highly correlated factors. Downstream party D has no visibility that the specific uncertainty emanating from party A upstream is 100% correlated with their own risks, along with parties B and C through which the liability passed on the way to them. What looks like independent risks are actually all the same risk and completely fragile to a single event that has quite a high probability of occurring.
> Downstream party D has no visibility that the specific uncertainty emanating from party A upstream is 100% correlated with their own risks
That's an exposure in the Party A-Party D frame is it not?
- Many underlying obligations are backed by insanely profitable companies.
- Financing involves long-term leases rather than short-term funding.
- Rating agencies analyze these off-balance-sheet obligations.
[1] it is an investment because I think xAI agreed to sell their compute infrastructure well below market price.
If OpenAI does not buy compute from Azure will they go bankrupt? Probably not - and that should be a . There is very likely a lot more discretion in these agreements than this article implies.
This sort of reporting is a clear indication of where sentiment is.
We (as a country) should contain our exposure to this insanity. If companies want to go bankrupt let them go, but we cannot strangle the liquidity of the entire market to feed the ambitions of these handful of companies.
Look at Russia. Nabullina was handed a complete shit sandwich deluxe with the war, but she did the economically right thing with the cards dealt, raised interests to 20%, forced businesses to trade in rubles, freeze equity/bond trading, cut regulation on banks, etc.
And the Russian economy which should have been in shambles like Irans because of sanctions is still kicking after soon 5 years of war.
Sure it's not really prospering and growing, but it's a graceful decay rather than a collapse and internal war like in Venezuela.
Even if AI completely busts and all of these servers are worthless that doesn't mean the data centers are. There's still a need for data centers to host non-AI servers. The hyperscalers will close/sell other data centers and serve from places where they have long term commitments at.
3.2 trillion sounds like a lot until you spread it over 20 years and half a dozen trillion+ dollar companies. It's hard to tell how much a bag they're holding but it's almost certainly negligible compared to revenue and market cap. The only real systemic risk is the risk of for some reason not needing any data centers.
If some lenders go bankrupt on a datacenter note post-buildout and the capital structure is reworked, lowering the costs by removing capitalization costs from the cashflow of the inference provider, that is all to the good of people who did not do the initial lending.
I think that is extremely unlikely, but it's good for datacenter doomers to think through these scenarios IMO.
I am not sure whether the dot-com boom was a net-positive, and the AI boom may be less productive.
Capital waste is only capital waste if the physical goods that were built are no longer useful for anyone. Otherwise, it's only a waste for the people who lent money.
For reference this 'massive build-out' resulted in, according to a Berkeley study I found, 500MW of total datacenter power in the US across 320 locations at end of 2020. This wasn't too much for very long: as of 2005, according to Tier1 there were 511 datacenters. And it was likely only really a year or two: Dupont's VA3 utilization numbers published it at 50% utilization in 2003, 89.4% in 2004, then full thereafter.
Meanwhile, these infrastructure improvements allowed, for example, things like the iPhone to be built and deployable -- there was 3g, there were a LOT more web devs that could create the early (web only) apps the iPhone needed to compete. This in turn allowed another round of investment to be built on top of all of that.
On the capital side, investors were also able to buy Amazon at around $1 / share. AAPL at iPhone launch was $7/share. I haven't split adjusted either of these, this is just from memory.
I don't think a bust was necessarily good but the boom undeniably built out infrastructure, a small fraction of what the world is currently using, and did it pretty quickly, and then post bust that infra got bought out in a capital efficient manner. I think that's the bear case here.
Put another way, the Dotcom one boom was wrong about two things: timing was too fast, and the predicted scale of change was way, way, way too small. They undersold the future.
They were underpants gnomes who had no realistic way of making money.
In that scenario the GPUs are a loss. But really at this point inference is so profitable it's hard to imagine these data centers not being able to at least cover operations & rent.
Like, trains are an incredible technology which transformed the world, but very few people go around counting their lucky stars that they live in a world that permits such wonders. It's a train, it gets things where they need to go, and by the way isn't it so annoying that sometimes it doesn't quite arrive when the schedule said it would?
That would be called a loan, I think?
If I commit to buying some future goods, and sign that I will hand over that money, that's debt. It's money I legally owe someone else. Counting that as debt is totally fair, not counting it is shady as fuck.
If I rent a house, I receive no money, in fact I receive no tangible goods at all really. Just the right to occupy a space under some pre-agreed rules. Yet, I now have debt.
How is this not a systemic risk? Asserting that:
> There's still a need for data centers to host non-AI servers.
Isn't true because a NVL72 rack consumes 23x the power of a traditional server rack which is why there's new construction. You could retrofit such a datacenter with normal servers but then you're taking a loss on the GPUs which is the majority of the cost.
At this point the question isn't really if we're going to use the GPUs they're building out. The question is how much profit they are going to make. I've seen people show margins as high as 80% on inference. That's a lot of headroom before we see data centers being shut down.
Sure 2.33T is less than 3.2T but these RPOs were $0.5T in 2022, and have been growing YoY at rates of anything from 84% (Microsoft) to 390% (Google.) Microsoft and Google alone are expected to recognize about $0.5T of that revenue over the next 12 - 24 months.
Is the CapEx and debt ("hidden" or otherwise) growing faster than the RPOs? Eyeballing the numbers, maybe? But generally service providers try to stay ahead of the demand curve rather than behind, so...
Is there concentration risk in the customers? Yes, OpenAI and Anthropic seem to be like $1T of these RPO's? That still leaves $1.3T of other demand.
Is there demand? I mean, RPOs come with a commitment, and have you seen the cloud revenue growth numbers recently??
As another comment here points out, is it really wise to assume the tech giants are NOT financially savvy enough to know what they're doing? Some of these, especially Amazon and Microsoft and yes even Alphabet, are well-known for being highly fiscally disciplined. This is a huge bet for them, even an existential bet, but maybe it's because this is an existential kinda situation.
Ray Dalio just found his next topic of conversation. The debt burden of the average AI agent. Those agents better get PRODUCING. GO WRITE ME SOME CODE CLAUDE!
In the higher interest rate environment we’re heading into this is a potential disaster. For the new players like Anthropic and their likely “junk” credit rating this is going to be incredibly painful. It’s basically like playing in a casino on margin. If it works, great. If not you are totally f!*ked.
Sure, smartest CFO's on the planet together with CEO's just collectively buying vaporware. Most of them have access to best metrics on the consumer side, and internals on where the industry will be in 1-2 years.
Maybe, just maybe, they can see how AI will eat the world, the same way software once did but much faster, hence they need these investments. what a crazy theory, right?
Just think about it for one second.. What if... AI that just change professions of people in Math, software dev, etc. Will keep changing any knowledge work and by doing that will accumulate a lot of cash. What if?
But I don't think they see it in ideological terms.
For them, the bottom line is that it's basically going to double cloud revenue over time.
How many times in history do you get to literally increase an existing revenue stream by 100%?
That's exactly what the zeitgeist was in 2007. The "smartest" guys were loading up on collateralized debt obligations. They were the sure thing. The music was never, ever going to stop.
Something you need to realize is that the "smartest" in our society aren't really the smartest. They are members of a group that exists on inertia and a desire to maintain an egotistical and social incumbency. They rarely suffer real consequences, least of all from their own decisions and actions.
These are the same kinds of people who looked at Elizabeth Holmes' business idea and were going to have your health decisions guided by someone who couldn't finish a bachelor's degree in biochemistry, or any other bachelor's degree. They were in tech, they were in academia (though apparently not at the University of Missouri [0]), they were in finance, they were in media. These phenomenons build upon themselves because it's a bunch of people in a club who can't admit to each other that the lemonade is really piss.
When there are questions being posed by people who are actually informed, who don't have a vested interest, those questions need to be answered. Even if the answer is "piss".
[0]https://www.darkdaily.com/2019/07/15/skeptical-missouri-path...
This another one of those times where Apple will need to roll up its sleeves again and take another component (memory) in-house there is no getting around it.
Is it possible that these companies feel they have more to lose by missing out on AI than by taking on debt? Or that they genuinely are in a race to superintelligence, which easily justifies this debt and far more?
When I worked at Apple, I remember seeing the commentariat insisting that <project> was bound to fail, that we didn't know what we were doing. When in fact extremely competent people knew precisely what they were doing. Of course, the commentariat were wrong and I'd wager most of them use what I had a part in building. The same story tended to repeat across my subsequent jobs in big tech. All this seems quite similar to me.
But current era also has echoes of capex financing (eg cisco lending to its buyers) and Enron (eg use of SPVs).
I don’t think that most people who are concerned about a bubble think people and orgs fueling it are stupid, but they’re operating with local incentive structures that can result is systematic problems.
And for folks who have lived through multiple of these (and maybe lost paper wealth, retirement savings, or homes), I think being concerned is reasonable.
Hard won lessons and all that.
That people don't do this, makes me think that it is more posturing and venting than an actual belief. People are uncertain about what will happen, but are angry and feel a lack of agency, and perhaps some FOMO as well. This becomes reflexive denialism.
I think AI is huge and it’s going to make a lot of money. So I’m keeping my investments in. But I’m also sure there is going to be a massive amount of turbulence to get there. Unfortunately I don’t know enough (or care to know enough) about finance to really capitalize on it.
* Previous wealth required.
Is this comment satire? Do you just sit in that weird ass Apple donut building, drifting further and further out of touch with real humans? Or do you leave it some times and interact with your average citizen now and then?
It’s just some nameless regular guy crunching some numbers and feeding to another regular guy who interprets the numbers. And so on
If it's just regular guys, then it would be arrogant for the HN regular guy to assume they can make better decisions with less information.
In reality it's not regular guys, these companies are very selective and have a very high talent bar compared to the median member of the commentariat. They also hire within domain expertise, whereas most of the commentariat is not of the same domain.
So in addition to an inside information disadvantage, the HN guy is also often at a talent and domain-knowledge disadvantage (as hard as it may be for the ego to accept this). This makes the confident contrarianism even more confusing, not even considering all the times it's been wrong.
Year after year of the advent of LLM's has repeatedly shown HN overall consensus gettings things wrong again and again. AI lab employees aren't here they're on Twitter/lesswrong. 6 months ago Mythos was announced and it has been an absolute deluge of "it's just marketing stunts." Practically 75% of many threads had that take. Strange, we don't hear much about that anymore, now that the plethora of hacks has taken place and increasing. There will be no mea culpa, just moving on to the next kneejerk dismissal as peoples' subconscious fearfully grapples with our immediate future. They want the AI boom to fail, because the implications of what happens next when it doesn't fail is unknown and vaguely frightening to them, for now.
Lots of people here called Mythos a marketing stunt, yet continue on in HN today having totally forgotten their original takes. I would bet some of them are reading this very post.
They don't even acknowledge that they were wrong, yet continue with the same behavior (reflexive contrarianism and denial) without updating. This confuses me greatly: why would anyone choose to be this stubborn, when it results in worse correctness? What do these people gain out of it?
(If anyone reading this fits the description - I'm genuinely curious.)
Since when did any employee ever do that? "It is difficult to get a man to understand something when his salary depends on his not understanding it". The higher up the food chain often relates in great part to accepting anything thrown at you with a nod and a smile. 6-7 figure incomes don't arise from sounding the alarm and creating problems for everyone within range
Also many of us are working in these AI / data center companies.
I can make any field of experts look like clowns by cherry picking their biggest fumbles.