Despite the AI Debt Pileup, Bond Market Still in La-La-Land, Investors Chasing Yield, Backed by Sky

With the stunning amount of capital getting sucked up while the AI revenue model is in fantasy land, you’d think it would be Panic City. But no. By Wolf Richter for WOLF STREET. The AI infrastructure investment boom is sucking up capital globally, both debt and equity capital, with huge debt sales, share sales, fund raises, and IPOs. The biggest IPO so far was SpaceX. OpenAI and Anthropic are on the horizon. But the IPO of Nvidia-backed Firmus in Australia, seeking to raise about US$5.5 billion, collapsed and was pulled. IPOs at sky-high valuations raise lots of money for the companies – $86 billion for SpaceX – and allow them to raise even more money later with follow-on offerings. This is capital that companies will never have to pay interest on or pay back. But that’s not enough, and so SpaceX has been in talks with Apollo Capital Management and some banks to borrow $40 billion to purchase GPUs from Nvidia, according to sources cited by CNBC two days ago. SpaceX already borrowed $25 billion in June via a bond sale, spread across several maturities, and investors stood in line to lend: that bond sale received $90 billion in orders. Yields were higher than Treasuries, but not much, and have since then surged, and spreads have widened. For example, the $6 billion slice of 10-year senior unsecured notes was sold with a coupon interest of 5.875%, so this was tempting, being nearly 120 basis points above the 10-year Treasury yield of about 4.7% at the time. The notes were rated at the low end of investment grade (BBB+ by Fitch and BBB by S&P Global; junk starts at BB+, see our cheat sheet for corporate credit ratings by ratings agency). Today, the 10-year Treasury yield is at 5.25%, and that SpaceX note trades at 90 cents on the dollar with a yield of 7.20%, according to TradingView. The spread to Treasuries has widened to nearly 200 basis points. So that caused a lot of handwringing because 190 basis points is the spread now between BB-rated junk bonds and Treasuries. And SpaceX is trading in that BB-rated spectrum now, like a junk bond. So now comes the next $40 billion in debt, but this time, SpaceX appears to target loans via private credit and banks, rather than the bond market. The bond market is getting saturated. In the US, there are about $12 trillion in corporate bonds outstanding, both investment-grade and high-yield bonds, according to SIFMA. The big AI companies alone are throwing another $500 billion at it this year. So some of the borrowing has shifted off balance sheet, to SPVs, to commercial leases, to purchase commitments, etc. And private credit is elbowing in on it. Then there are other big borrowers, such as the Paramount Skydance acquisition of Warner Bros. Discovery, that is funded by $52 billion in debt sales, including $44 billion of bonds, of which $12 billion are junk bonds, leaving the combined company with about $80 billion in debt. It’s not like the rest of the world is waiting for AI to get through borrowing. The debt markets globally, private debt firms, banks, and other lenders are struggling to absorb the huge amounts of debt capital that the AI firms are in the process of raising, or plan to raise, in addition to the debt sold by other companies, and good lordy, by governments. The US government alone has to fund about $2 trillion in deficits a year. AI debt is risky. The amounts are gigantic. The revenue models of AI companies remain in fantasy land, as neither consumers nor businesses may ever be able to spend enough on AI to justify the planned AI capital expenditures. You’d think it would be panic city. But no. Yields have risen, but are now essentially just back in the pre-QE normal range that prevailed before QE was instituted across the world in 2008 and after. And spreads of corporate bonds to Treasuries have widened, and for some specific bond issues, such as the SpaceX debt, have widened a little further, but they remain narrow by historical standards, as investors are still chasing yields. The spread between BB-rated bonds – upper level of junk bonds – and Treasuries of equivalent maturity widened in early October to 2.04 percentage points, then narrowed again to 1.94 percentage points currently, as per the ICE BofA BB US High Yield Index Option-Adjusted Spread. That spread is still hobbling along historic lows. In March 2020, the spread went over 8 percentage points. During the Financial Crisis, the spread topped out at nearly 15 percentage points. Even following Trump’s “Liberation Day” speech on April 2, 2025, the spread went over three percentage points. These narrow spreads at the upper level of junk bond land – despite all the handwringing out there about widening spreads – show that the corporate bond market is still in la-la-land. Part of that is driven by the AI-related stocks with trillion-dollar-plus valuations and by future trillion-dollar-plus IPOs such as by OpenAI and Anthropic that allow companies to raise lots of money with follow-on share offerings from the huge and deep global stock market. that’s confidence inspiring for the bond market. But when those shares plunge, making share sales difficult or impossible, that’s when the bond market would get frazzled. The investment-grade senior unsecured 5.875% 10-year notes that SpaceX sold in June at a spread of about 110 basis points, are now trading like junk bonds in the BB-range (upper level of junk bonds), depicted in the chart above, with a spread of nearly 200 basis points and a yield of 7.2%. So yes, investors are taking some losses on these AI bonds in terms of lower prices due to higher yields and wider spreads, but yields are not high by pre-QE standards and spreads are still historically low. None of these AI bonds or loans have defaulted yet. All the plates are still spinning nicely. Sure, the Firmus IPO got pulled as investors balked at the valuation and maybe because they finally glanced at a prospectus, but that was an Australian IPO that really tried to push it. The B-rated spread – that’s the middle of junk bond land – has widened just a tiny bit to 3.15 percentage points, but also remains historically narrow. This still has la-la-land written all over it. Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the mug to find out how: What if yields go to 15 – 18%, Wolf? Then I’ll be able to buy you a Yummy Burger with my yield earning! 🍔 The bond market is a-tremoring. 1% at a time. It went barely over 5%. a BRILLIANT presentation. I believe nearly 1/2 AI data centers are getting huge road blocks or like oracle are shipping in bottled LNG to run generators as main lines of NG are being delayed and/or re-routed only costs 4x with some 90 trucks per day needed If the revenue models of these AI companies are fantasy then why do you think they are doing it? What is their end game based on your experience? These are not dumb people. How could it be that so many smart people seem to think that this will pay off? Winner takes all, everyone else loses? Not in America. If you are big enough, say a SIFI bank, or automaker, or a large insurance company like AIG with derivatives counterparties in banks, or maybe just a foolish student who borrowed from the government……you cannot fail. Or maybe you are a crony capitalist with connections you lobbied for. Or how about you make something or do something foe the military, you are too strategic to fail. Better yet, Trump picks a new class of winners and sends them money on a daily basis. AI has already let the cat out of the bag; they NEED regulatory capture and oligopoly profits. And think, what will the 585 buffoons in Congress do when retirees whine their 401ks are losing money in failed AI business models. Wolf talks of winner takes all; which is still, if allowed to happen, a byproduct of capitalism. The cannot let many businesses , industries, investment fail is socialism. The 70% of our population with no investment assets are getting slaughtered in this strange game. That’s just nonsense. We’re talking about INVESTORS getting bailed out. The only examples where investors got bailed out were the banks at the peak of the Financial Crisis. Unsecured bondholders — that’s what we’re talking about here in the comment — get wiped out routinely, no problem. Secured bondholders get the collateral, whatever that may be worth, and that’s all they get. GM, Chrysler, Lehman, many others went bankrupt, and investors lost everything or nearly everything. Huge component makers went bankrupt, shut their factories in the US, and investors lost everything. Ford didn’t go bankrupt, but wasn’t bailed out either. Bear Stearns was scuttled and sold for scrap and investors lost nearly everything. Lots of huge companies went bankrupt, and some were liquidated. SVB and First Republic were torn apart and sold for scrap and their investors lost everything. There are thousands of examples like that where investors lost everything or nearly everything. At the peak of the financial crisis, the big banks were bailed out and investors were made whole, and that was the exception because it was a financial crisis and those were banks, and regulators didn’t want the global financial system to collapse. And since then, no investors got bailed out. “There are thousands of examples like that where investors lost everything or nearly everything. At the peak of the financial crisis, the big banks were bailed out and investors were made whole” 7,000,000 homeowners weren’t “made whole”. Where is the wall st. bankster that went to jail for fraud? War is a racket and so is Wall St. The biggest regulatory moat is around Wall St. If every household subscribes to AI for $20 a month, that’s $3.3 billion in revenues a month, or about $40 billion in revenues a year. That’s not going to pay for even the paper that the multi-trillion — with a T — investment bubble is built on. > If every household subscribes to AI for $20 a month… That’s not the revenue model they have in mind. Think of every software engineer paying (indirectly, through their employees) their own salary in tokens. That’s possible and is happening because AI more than doubles their productivity. Soon enough, AI will 10x their productivity, and the same will happen (or is happening) with mathematicians, economists, journalists, managers… foo That’s at best a cost-saving model, not a revenue model. Saving costs means slashing someone else’s revenues that used to provide this service. SO if AI entails big job losses (the cost-saving model) and cost savings in other areas, it will entail a decline of corporate revenues and profits — because those unemployed consumers and companies that lost their revenues aren’t going to spend much anymore and — a decline in tax revenues, a collapse in stock prices, and bond investors can kiss their money goodbye. All this stuff you cited in your comments is financially illiterate or just propaganda because you don’t understand the fundamental dynamics of an economy where companies can thrive. If you try to run an economy on cost savings, you crash that economy. An economy grows when spending grows. What AI needs to do is to create ADDITIONAL revenues in vast amounts in the overall economy, including from consumers with lots of money to spend, and from businesses with lots of money to spend and invest, and if that doesn’t happen, there is no revenue model. That revenue model for AI still needs to be found. heads they win, tails the banks and suckers lose. just like commercial properties original developers. and countless episodes in the history of capital markets. the best book on this subject, is “this time it’s different, 600 years history………” Well AI is a prime example of the lure of the perpetual motion machine. temperature in pot about to hit ‘HOT’ frog fat and happy as I feed him another ‘grasshopper’ Definite losers. Right now, despite the intentional fog about everything AI related in media, AI technologies are domain specific. Generalized AI is not on the horizon and there are strong arguments pro/con about whether super intelligence can ever be achieved. So the competition is within these specific domain areas (application areas). Or so the argument goes. In a word, fees. The end game is the IPO into this wild bull market on top of the insane bubble valuations. They will probably raise few hundred billion dollars between OpenAI and Anthropic. Then they will buyback owners/executives shares at these same valuations. Then après nous, le déluge. These hundreds of billions of dollars will go directly to the semiconductor industry. They are the big winners of the current phase of the AI buildup. I would not use the phrase ‘go directly’ to describe circular financing schemes. foo Seems you don’t understand where this money is coming from. Namely from investors that bought the bonds, loans, and commitments… that’s what the whole article is about. And when that stops because investors that handed over their cash and bought the bonds and loans and commitments that paid for those semiconductors refuse to risk even more? And sales of semis collapse? And then semi prices collapse because demand has collapsed? Semis go through horrendous cycles. That is what I think. That there are no winners, in America today, there are only losers. A gas bag deflating. What are their revenue models? I agree they are smart people. Seems like they are in the give it away mode right now. Is the chatbot data that valuable? “Is the chatbot data that valuable?” Yes. The fees from end users is not where the big revenue comes from. The big revenue comes from the spreads between what people can/will pay vs the cost to produce. The detailed, personalized information that AI can gather (and sell) can allow sellers to significantly widen those spreads. It’s a paradox as to whether or not this should be called an efficient economy or an inefficient economy, because it is maximally efficient for the sellers and maximally inefficient for the buyers when the data is shared, and There Is No Alternative for the buyers because the government refuses to step in and break monopolies/collusion. PS A subexample of buyer/seller is datacenter/software-company. If I have to pay $50,000 for $4,000 worth of server time to create a product in a week that would cost one year of programmer man hours over six months, say $160,000 in programmer costs, I would take that trade off in a heartbeat. The datacenters and software companies are both huge winners here — the programmers are the only losers. The PROBLEM here is that there are literally thousands of essentially free Chinese open models that are producing better results with each passing day, and more efficiently. Likely, within about seven years, the software company will be able to buy it’s own hardware for $50,000 cutting the data center out of the equation entirely. And that $50,000 can produce a lot of software until it becomes obsolete in three years and needs an upgrade. My take? Another Bronze Age Collapse is coming because this is a train no government can stop, no matter how hard they try. The genie is out of the bottle. Yeah, you have to assume that the customer data is the product. The thing is, how much more data do they need on everyone? Is the value of ever more data ever reaching a diminishing return? JeffD….I enjoy reading your thoughts on all of this. Question: do you believe much of this conversation is overly dominated by interactions with LLMs and ignoring other AI application areas? There will be markets in these other application areas (or domains as I like to label them) – there already are (e.g., self driving cars). I don’t pretend to fathom how all the plusses and minuses will ultimately sum…but I do know there are valid AI application areas that people, businesses, governments will pay for. @Phillip Jeffreys, My bottom line on AI is that at some point, as the adoption rate accelerates, we are going to see wild swings in money flows, vs the relatively smooth behavior we are still seeing. As more income gets diverted to AI’s “income consumption”, it’s going to get diverted away from competitive areas of “normal consumption”, causing unpredictable and widespread dislocations in the economy. We are only in the first inning, and we can already begin to see the effects! I guess I’m saying I can’t predict how things will evolve, I only know that there will be shocks similar to those created during the industrial revolution. The other comment that comes to mind, more in line with your question, is something I have commented on in this forum before. I think the software tied AI will make the white collar work more productive, and that the hardware tied AI (aka industrial robots) will make some of the blue collar work more productive. The white collar disruption will be larger, (1) because the productivity there can evolve faster than in the blue collar work and (2) more aggregate income flow is tied to white collar work (services) than blue collar work (goods). That said, the above comments stray away from my area of personal expertise before I retired, which was (1) how to optimize computer systems by appropriately mapping which part of the problem software should be responsible for vs hardware and (2) how to specialize general mathematical techiques applied to systems into more specialized/optimized mathematical subsystems for each part of the problem. The revenue model is not fantasy, at least for the companies that sell semiconductors. The hyperscalers make money from advertising and other revenue streams; and all that money is transferred to those who sell chips. As long as Google, Amazon, Microsoft and Meta have other sources of revenue, the infrastructure buildup will continue. a very insightful point. hat tip for pointing that out. just got back from an AI conference of big boy investors and tech analysts………..it’s a bubble of course. but like the tech and r/e bubble it is best to ride her like a nice wave in rockaway beach and bail out when the tide turns……….worked for me in 2000 and 2008. and 1998 LTCM crash and burn. What you describe is NOT INVESTMENT if there is no pay out in the end. It’s just throwing more good money after bad money and obviously that will not continue for long. This is borrowed money we’re talking about here (bonds, loans, commitments, etc.) and someone has to ultimately pay for it, or else these investors whose money this is are going to have to say goodbye to their trillions. That’s what this is about. Yes, these companies prefer to issue bonds because that’s a better allocation of resources. Why would they invest their own money if they can borrow with a relatively low interest and have much higher returns? NB: As a software engineer, I can attest the *demand* for these services. I’m not talking about $20/month, but thousands of dollars per month (per knowledge worker). Where did all that money being invested in AI come from ? I suspect it is coming out of the pockets of all the people who benefited from the Covid giveaways. I figure they can afford to gamble it away and lose it. Poof, it is all gone. Well, maybe it was converted into stuff nobody wants or will use. It could even turn out to be deflationary. Wolf, if you have the time, I am wondering what the rising interest rates are going to do to all the “Zombie Companies” who borrowed money at ZIRP rates to buy back their stocks ? That should be an interesting topic for your viewers. They’re acting as if the first person to AGI takes everything. But AGI is a big bet that I don’t believe has any chance of panning out. However as long as investors are willing to throw money at it then somebody has to chase it. Alternately, they may be chasing the Amazon / Google winner takes all scenario. I watch a lot of commentary on this and what nobody understands is how that works in a situation where you don’t really have a moat. Granted, say, Google search also doesn’t have an intrinsic moat, but it also doesn’t cost anyone anything to use. AI currently costs money and will have to cost a lot more money in the future, and that creates an incentive for people to seek the cheapest model. Already I’m hearing about companies setting up “routers” which look at the prompt and decide which model to send it to that’s most cost-effective. As AI labs are being forced to switch to token-based billing, that means that nobody has an incentive to stick with one company because they’re not paying a subscription. The incentive structure here has become like an anti-moat, so nobody knows how this is going to work. Fund managers meanwhile have a perverse incentive to buy this stuff even if they know it’s a bubble because “a bubble is something that I get fired for not owning.” One guy I watch calls it the “who could have known?” phenomenon. “They’re acting as if the first person to AGI takes everything.” Yeah, this is the part I don’t get. I switch back and forth among GPT, Claude, Gemini. Whoever reaches AGI or “wins” as a certain president puts it, BFD. It’s not likely that’s the forever winner. Can’t imagine there’s much of a moat. Please stop you are making Claude uncomfortable like a human child. Has anyone asked AI how this would all turn out in the end? The real question is: How much are you willing to PAY AI for the answer? That’s part of the revenue model… you’re getting the answer for free! And if you use a browser that blocks the ads, you don’t even see the ads. The investors paid for that, and are getting zero in return. Trying to get consumers to pay for AI as much as they pay for groceries will be tough. I pay $20/month for ChatGPT. The free model usage limit is just too small for my needs. Probably best $20 I ever spent. I also pay $20/month for Anthropic’s Claude but will cut it soon. Claude is good for some things, but ChatGPT is good for most things -research, documents, planning, you name it. Google AI is just retarded compared to the above two. How much are you spending for food a month, and are you willing to spend that much on AI? That was the question. $20 a month is not in the ballpark to make these huge investments work. $20 a month is a lot less than a Netflix subscription. Its half of a WSJ subscription. It’s nada. I can see myself paying $60/m for a useful AI model that is very productive, maybe a $100/m. But they are offering all I need and more for $20/m (for now). However, I can also see how “addictive” a good AI can be. Once you get used to working with AI, things just seem imposibly difficult without AI later. So there may be some pricing leverege there. If I had to guess, everyone will be subscribed to AI in not too distant future. That still does not even remotely justify $6 Trilion value of Nvidia and the like. We shall see. The real money is not in the personal use of AI; at least for now. Companies are losing money with domestic users because they expect people will bring the same tools to their jobs. That’s where the real money is: work. Let’s assume an average Software Engineer makes $150k/year. If AI can double their productivity (and it can), it is worth $150k/year. That’s why large companies are already allowing some engineers to spend their entire salary with tokens. If AI can 10x their productivity (it will, in the near future), it will be worth $1.5m/year. I don’t know which company will receive the monthly bills; but I know one thing: AI will require chips. The semiconductor industry is the big winner. foo, AI is a brute-force approach to programming (and everything else). It works while they are throwing billions of dollars at it. Of course, it generates some spaghetti code when you use ten million times more computing power compared to a regular engineer. You can break 256-bit encryption with enough brute force. None of this means it will be cost-effective. And a lot of Genz are completely against AI. They want a better world not shackled to these companies. Always an enjoyable read, thanks Wolf. If the general Ai becomes free, I could see a business model being to harvest all the good ideas that people type in and get out ahead of them. A good idea for an invention, found a loop hole to Black-Scholes model, novel block chain code,… ppl will ask Ai its opinion and for help. Why wouldnt the companies troll these ideas and get out of ahead? I digress, but that’s probably the only issue with the matrix, movie, lol, humans as batteries? Silly. Warehouses of humans hooked up in parallel to produce ideas for the machines. Priceless. We’re the billions of Nvidia chips in this scenario. > That’s part of the revenue model… you’re getting the answer for free! Software Engineers in large companies are spending $1000+ per month in tokens. Some of them are allowed to spend their entire salary in AI. Not only it pays for itself in terms of productivity, but, as we are starting to see in mathematics, humans with AI are solving previously unsolvable problems. foo And those token prices are already collapsing, and those budgets for tokens are already getting curtailed. Because there is no endless magic money. I never did, but I have thought about it. Here is the response from my el-cheapo free version of Copilot: ———————————————————————————————- Where the bubble does exist Certain parts of the AI ecosystem are clearly inflated: Startup valuations Companies with no revenue are being valued at billions. That’s bubble behavior. GPU demand NVIDIA’s dominance is real, but some companies are hoarding GPUs without a clear plan to monetize them. Corporate FOMO Many businesses are spending heavily on AI because they’re afraid of being left behind, not because they have a strategy. Promises of AGI Some predictions about “superintelligence next year” are hype-driven and not grounded in current research. These areas could deflate—hard. The non-obvious insight The biggest risk isn’t that AI is a bubble. The biggest risk is that AI is a winner‑take‑all market. If only a handful of companies (OpenAI, Google, Anthropic, Meta, maybe Apple) end up controlling the entire stack—models, chips, data, distribution—then: Many startups will die. Many “AI-powered” companies will be commoditized. Investors will lose money even though the technology succeeds. This is exactly what happened with the internet: the tech changed the world, but only a few companies captured most of the value. My position AI is not a bubble in the sense that it will “pop and disappear.” AI is a transformative technology with bubble-like behavior around it. The tech will survive. Some companies won’t. Harvey Mushman aka Steve McQueen. What is described there always happens to new technologies. Back in the 19th C there was loads of small local railway companies that eventually consolidated into a few. Also refer to 20th C automobile companies. Only difference now is that these processes happen a lot quicker. AI companies plot ‘day after’ scenarios for public revolt… ANTHROPIC Model Goes Rogue, Submits Fake Unsolved Murder Tip… The bond market is getting saturated. In the US, there are about $12 billion in corporate bonds outstanding, both investment-grade and high-yield bonds, according to SIFMA. $12 trillion? But there’s still demand. Yields are increasing, but mush higher above pre QE. Go Bills Yes, thanks If, as it seems to be the case, everyone can use AI to make an entertaining feature length movie in one day, where will Skydance’s customers come from? We will go back to live plays. Broadway. Why? Because humans are human and we enjoy humans of talent, human community and human interaction So, when it hits the fan, will the Fed end up buying AI backed bonds? Adding more junk to the balance sheet? No. It will let the market sort it out. Markets are good at sorting this stuff out. i know i do not know. but i remember reading twice with my pal from jane street, the “audit” of bailout in 2007 and 2008……..that bernie and ron paul commissioned. the us gov bailed out dumb hedge funds in europe. i would bet in future years they would quite easily bail out the us domestic AI companies. i’d bet a gold coin and steak dinner on that. i tried that bet already with some smart pals. no takers. A plausible scenario would be a government “investment” in failed AI companies, similar to the Trump admin’s action with INTC. But those AI dudes are smart and would likely seek that “partnership” well before actually failing. I don’t know what all the Fed will do, but I’m convinced that they’re not going to sit back and do nothing. Upwards of $3T, whether mostly from private investors or not, is going to serve as a very big drag on the economy. Private investors losing money has no impact on the banking system, and so be it. Nobody cares. When the big tech and AI companies drop by 30%, that’ll wipe out something like $5 trillion, and nothing bad happens, just paper profits going back where they came from. The issue for the Fed in 2007-2009 was the banking system – it was on the verge of collapse because it was loaded up with $10 trillion home mortgages – and that was a lot of money back then, US nominal GDP was $14 billion back then — and those mortgages were going bad, and several bigger banks, mortgage lenders, and investment banks had already collapsed and were being liquidated. The Fed feared that the entire financial system would come down because of those mortgages. Now the government guarantees the vast majority of mortgages, and banks are off the hook. Most of the AI debt is held by investors, not banks. Spinning plates is a great analogy. Hope the splash from those plates tumbling down is minimized but not holding my breath. “The revenue models of AI companies remain in fantasy land, as neither consumers nor businesses may ever be able to spend enough on AI to justify the planned AI capital expenditures.” When I read the above quote, the first analogy that came to mind was California high speed rail. So just how much would each one-way ticket have to cost to “break even” on the high speed rail project? According to Gemini AI: “If a theoretical cohort of 30 million total passengers paid entirely toward capital recovery, every single ticket would need a surcharge of $400 purely for the construction cost, on top of normal operating and maintenance (O&M) expenses.” I don’t agree. That sounds too cheap. No one can do basic back of the envelope math anymore. All that matters is “GDP growth” aka bigger loans for failed enterprises. Extend and Pretend to the moon. Maybe there are parallels but AI software does not seem to provide the same benefits as core infrastructure, as such its payment model should be completely private between producing company and user. Rail…you don’t need to pay back the capital in rider fees. There is a big economic value in a mobile population. That increases economic output which provides the tax revenue to pay for the project. And to be clear I’m not saying CA high speed rail is cost justified. It’s just different than an AI company selling its software capability The biggest risk to AI is that they have overbuilt vs demand. The technology used in current buildouts will probably be obsolete in three years. Running on current servers may be 10x as expensive to maintain as the next generation. PS “overbuilt vs demand” — same problem with California high speed rail, BTW. The ridership projections for high speed rail at one point was ~110,000 per day, and still is to my knowledge. Compare that to the less than 20,000 people on airplanes each day between northern and southern california, including all flights from all airports, going both directions. “Trading in the iShares 20+ Year Treasury Bond ETF (TLT) Wednesday was very call-heavy on volume that was 50% above the 30-day average, according to data compiled from Cboe LiveVol and SpotGamma. Traders bought almost 370,000 calls versus under 100,000 puts in TLT, and sold more puts than they bought.” A little nibbling. Why is any of this GAMBLING SPECULATION allowed at all? Cuz dudes are gonna gamble. It’s in the dna Cleanup on aisle finance 📢 • Hyperscaler CapEx (Big Tech): Major tech companies (Amazon, Microsoft, Google, Meta, and Oracle) are projected to invest over $700 billion to $725 billion globally and domestically on AI infrastructure/Capex in 2026 alone. • Long-Term Projections: A Brookings Institution study estimates total U.S. AI data center investment will reach $10.3 trillion through 2032, while global estimates scale up to $31.6 trillion through 2050. AI is headed straight into the abyss, another too big too fail conundrum, these number make our $40 trillion national debt look like peanuts. The revenue streams by 2032 must be gigantic. “…reach $10.3 trillion through 2032,” You can take that off your worry list. It won’t last that long Few times I’ve seen you so certain when predicting something in the medium term, Wolf. Oddly comforting, in this case. Thus far it seems to me that the only thing AI will do for me is to give me access to information, which is a good thing. But unlike television, phones, electric bills, or whatever, AI information appears to be free. How the heck are they going to make any money. I don’t get it. It is almost as if they think that money will not matter someday, that all debt will be forgiven, and we will all live on fantasy island or whatever. Didn’t Elon Musk say that money will become obsolete? Let’s say you spend 3x what everyone else will pay on airline tickets, to purchase airline tickets using your credit cards, three times in a row. Let’s say that information gets shared among all airlines and travel agencies. Do you think you will ever be able to buy an airline ticket at the “going rate” ever again in your lifetime? Not bloody likely unless government regulators step in, and I don’t see them stepping in yet. Thanks Wolf……focusing on the elephant in the room that mkt refuses to acknowledge. And thanks for pointing out how much hard earned was lost in previous debacles, wiped out seemingly in a flash, with no hand outs from Washington. Still own rental properties, shouldn’t, but ignored info you presented….at considerable cost…… Paying more attention now…… Hoping that, after considerable pain, which we have yet to go through, Walsh will be forced to reflate……but, as they say, ‘hope’ is not a strategy…….but it’s my guess anyway. Between La La Land and the Drunken Sailors, it seems to me like there is still too much money in the system. It’s not money. It’s CREDIT and massively inflated asset valuations that there is way too much of in the financial system. Moral hazard induced la la land. I’m amazed how the VIX is sitting around 14-15 despite what’s going on. Is that a result of ‘global tracker’ positioning? What about the junk bonds index funds SPHY or JNK or HYG or HYD? This way I dont have to buy individual bonds but buy the broad market? Look at a 20-year chart of HYG. In 2007, it was $105; at the multi-year peak in 2013, it was $95; at the multi-year peak in Jan 2020, it was $88. Now it’s at $77. In between those peaks, it fell as low as $62. That’s a lot of risk and loss of principal. Back-of-the-envelope, the fully burdened cost of labor in the U.S. is $16T per year. I have seen estimates of AI replacing upwards of 25% of that labor, or $4T per year. So the AI industry is spending $0.5T in capital per year to capture a $4T market. Assuming AI costs pennies on the dollar to do the same work as human labor (and can work 24/7/365), that $4T drops straight to the bottom line. The ratio of $4T to $0.5T or 8X seems favorable as a capital investment. The key question would seem to be how long does one have to do the $0.5T per year capital investment to fully realize the $4T? And then what are the ongoing O&M costs? And how long does the hardware in an AI data center actually remain viable? I have heard no more than 5 years, and the compute hardware needs to be replaced. To what extent is true infrastructure being built, versus obsolete-from-the-get-go technology? There is also the cost of the workers who are no longer working, and that is perhaps $4T/1.5 = $2.7T, which would fall on the government to deal with. I guess part of the model here is that the AI-driven GDP growth would cover the public cost of living for the workers who are no longer working. And how much will it cost to tear down the data centers and do environmental remediation, when AI runs its course, ~ 25+ years from now? And who will pay? This has to be factored in, as well. Is there some canonical, sober, unbiased macroeconomic case for AI capital spending? Do any economics brainiacs have a handle on this? Or maybe AI could explain itself to us? Would we trust it? “AI replacing upwards of 25% of that labor, or $4T per year. So the AI industry is spending $0.5T in capital per year to capture a $4T market.” This is Exhibit A of how stupid and braindead the AI financial hype is. If 25% of the working people lose their jobs, WHO is going to pay for AI????? If you have a worse-than-Great-Depression labor crisis, how are investors ever going to make money? In your scenario, tax revenues will collapse, corporate revenues will collapse, stocks will collapse, bonds will collapse, AI funding and use will collapse, and leveraged investors will be wiped out and jump out of windows in groups. I stopped reading after that sentence. That BS is just too dumb, ignorant, and manipulative to pollute my brain with. Fair enough, but what then is the business case for AI, if it does not allow for replacing workers with presumably more efficient (lower cost) machines? I guess the argument is that everyone keeps working and getting paid the same or more in real wages and AI does what? Is AI another uncorrelated source of profit on top of the labor force? I would think labor would simply be replaced and fewer workers would be needed. Perhaps not 25% but some number >> 0%. Only IDIOTS invest in ANY BONDS or SIMILAR CAPITAL RAISING INSTRUMENTS these Days .. NOBODY needs ” A I ” … especially paying for that SPIN – JUNK ….. A I is a SCAM of IMMENSE PROPORTIONS … My 91 year old neighbour used to be the office manager of their business. She can still add an entire page of numbers of several places, quite rapidly, in her head. I used to fly bush planes in the north and can still visualise the 4 mile maps (4 miles to the inch) in my mind, and see for several hundred miles in all directions every valley and stream and mountain pass in 2d representation. Make a mistake map reading and you literally die. One pass had several wrecked airplanes on the dead end hanging valley (to the left) constantly reminding you. (Hell Roaring Creek south Nahanni River). Maps let you know exactly what you needed to know and remember and get the job done in less than one mile visibility. Today’s airline pilots cannot even fly their aircraft hands on as Korean Airlines SF Intl can attest. . Old time sailors knew trig and celestial navigation to the second. If they did not know this skill they died. My mom learned Latin in high School (public school). She wrote books until age 80 and finally lost her eyesight. Nowadays? People ask AI what to do and what to think….for some it is a friend and relationship. Sit in a waiting room or restaurant and watch brain dead people endlessly scrolling their phones, seldom even reading what they see. People, my relatives included, cannot drive anywhere without a GPS deciding for them where to go and when to turn. People listen to podcasts to learn their opinions, and these reactionary polarised times tell us exactly how well that is going. And now everything is AI. AI this, AI everything…… for the economy and investment and for the race to supremacy. This is not going to end well. The more electronics are used to think for humans, decide for them, manage them, make them adapt….the dumber people become. Devolution. It is ironic. Complex thinking machines writing their own code as an extension of a tool….a modern day hammer or winch of 0s and 1s. All this while people are less capable everyday, less able to buy a home, groceries, or afford healthcare. To think. At the risk of sounding like a conspiracy theorist on a fringe podcast, methinks these tech titans are really designing a race of puppets programmed to hand over their money and future. Willingly. Where are the Luddites when you need them? As a self acknowledged and confirmed old neo=luddite, I can only say that ”WE” are here and well and doing as much as we can within the clear restrictions imposed by the little green and other coloured folx following us around and trying to stop us… Other than that, we neo-luddites usually try to keep as low a profile as possible due to the clear and obvious GUVMINT and CORPORATE goons, etc., etc., Not going to reveal on here the various and sundry and extensive groups existing to maintain and continuing to try to extend the luddite actions, but you can find them on line fairly easily these days, especially if you wish to contribute… Good Luck and God Bless your every effort to hold back the tides, in spite of the total failures to do so over the last several millennia!!! in the 1980s i was a cartographer. we surveyed by hand a small harbor in maine for the USGS. i also worked for NASA and another gov task force using state of the art GIS and remote sensing and computer cartography hardware and software. i agree with you that the foundations are important. like your bush pilot days. At a glance, it appears to me the trouble can start here when ‘investors’ and I use that word very lightly, buy these bonds on borrowed money from something that lends cheeper than the bond yields deliver; thereby creating a positive spread for the ‘investor’. And the investor leverages up to the 9’s. Until the party stops because the underlying was just a science fiction ruse to begin with and the charade goes Enron 30. ding ding ding. At the end of the day its all about making money ‘legally’. Guys I work in Health Care tech which is typically more conservative. AI is absolutely collapsing timelines and resources to deliver to end users in one month what may have been and 6-12 month cycle with large teams. Now you can build in house what was shipped by external vendor where you had to fight the roadmap for your priorities. Product and Engineering roles are consolidating, claude gives massive contextual knowledge To developers. Analysts and admins can be running multiple files at once – backend configurations that took hours to build and test – done in under one minute and usually with much higher accuracy than the human after spending hours cross eyed. I would quit if AI went away. Executive facing PowerPoints that took hours and collaborated effort to refine- done in 10 mins or less. The Microsoft tenant knows your entire one drive emails teams meetings etc. And it gets significantly better each WEEK. Only a matter of time before I am redundant. The only think it can’t do for security reasons is peel the onion on data drift and go look do lookups in sql when the end user questions a report and you have to go check the ten source tables to figure out what failed. But an agent can be trained to do that with right access. I have been around the block and have never seen anything this consequential. There is so much grind and complexity in technology and the domain knowledge is highly specialized- AI is democratizing it AND getting 90% of the grind done for you. New roles are coming out is this. Some ROI is real some too muddied with other factors to be sure. FTE reduction is real. So what happens when we are all replaced with autonomous agents? This will be a cycle – when everything starts looking the same, talking same, working the same which is already happening they will hire back humans as the differentiator. Authenticity will become the Crown Jewels. Of what value is ANY of that stuff? It would be far better and cheaper just to eliminate all of that nonsense. The coward Jerome Powell was too yellow to do the right thing. All of this nonsense could have been halted years ago. Instead, we have the greatest waste of natural resources in world history. What was the ‘right thing’ that you wanted Jerome Powell to do? And do you not realize that he is just one of the 12 members of the FOMC which vote to determine Federal Reserve policies? I have some personal familiarity with CIO’s of medium sized enterprises who are enthusiastic about AI. These folks are conducting various trials and trying to get a firm handle on were the low hanging productivity fruit is within their businesses are. As those trials wrap up it feeds planning for their future direction. The consensus seems to be that AI is real, improves productivity but not of the “fire 50% of the employees” variety. They are also moving toward running open source/open weight models on their own compute – both because it’s cheaper and because they dont want their guts living outside the corporate body in someones data center. Havent we seen this whole AI picture before. Werent the Rialroad booms in the 1800’s or the Fiber Optic boom of late 1990’s-2000 similar. Revolutionary tech with lots of real productivity implications. Massive investment, massive capex and much of it ended up being wasted. Many of the early companies went bust and the real winners where those in the 2nd generation who bought valuable assets out of bankruptcy at big discounts. Railroads are now legacy. Its only now that a lot of that fiber capacity is getting utilized. I suspect AI will be the same deal but everything will happen much faster from boom to bust to boom again. “You’d think it would be panic city. But no.” What isn’t factored in any market is that the US government may well burn the entire system down on its way out. It’s been a long long time since anything like this was seen, but it’s not a first. There are kings whose names were tried to be erased given that’s what people had at that time. We are in for a real treat ahead and no market is going to work and in fact the Market system may collapse with it all. But we will survive as we have from past incidents like this. Two questions about the investment-grade senior unsecured 5.875% 10-year notes that SpaceX sold in June. (1) Why were they rated as investment grade? (2) Who on earth bought those unsecured bonds at that rate? That seems crazy. Good questions!!! musk, like trump has a cult. many are wealthy. I missed this in April, but apparently Amazon offered a 50 year corporate bond at 6%. 50 years! Wow Yes, it’s now trading at around 90 cents on the dollar. This article from Henrik Zeberg is quite interesting on the historic perspective of bubbles and why the technology can be transformative, but still it is an investment bubble. Close to what Wolf has been saying, but this has some historic perspective – railroads! The $20/month could be the reality after things settle down. One or more of them will find a way to 10-1000x the throughput/performance for a given set of hardware decreasing cost/cycle and increasing adoption – Jevon’s paradox. So those humungous data centers might be like the mad dash during the .com to build out Fiber optics but at a crazy larger scale. Just a thought.