Yes: AI Is a Bubble. But It Is a Bubble-Plus. & That Makes a Substantial Difference
Bubble-double-plus: how the AI boom props up the economy with debt, data centers, faith, and valuations without rational expectations of durable and monetizable productivity growth…
Bubble-double-plus: how the AI boom props up the economy with debt, data centers, faith, and valuations without rational expectations of durable and monetizable productivity growth…
The number of people uneasy about the flywheel between the macroeconomy, the stock market, and the ongoing AI bubble is growing. Here is Tracy Alloway:
Tracy Alloway: Here’s what Tracy is thinking about <https://www.bloomberg.com/news/newsletters/2025-10-29/some-thoughts-on-today-s-fed-decision>: ‘Just how central is the stock market to the overall economy? If wealthy people are driving spending and the assets at the heart of that wealth keep hitting new highs, then it would make sense that equity markets become an increasingly critical support for jobs and economic growth…. As stocks have notched new records, we’re seeing new products and strategies designed to help the wealthy tap into those gains…. People using box spread strategies to create a sort of synthetic loan at a lower rate than they might get from a bank… effectively borrowing against their portfolios…. Stelrix… pitches itself as the first credit card that enables users to automatically borrow against their stock portfolio…. The target market for this card — which apparently is “crafted in proprietary glass by the same minds behind the Black Amex Centurion” — seems pretty obvious…. The money’s not just getting spent as stocks go up, it’s getting multiplied…
Yes: the AI-bubble is pushing up investment spending directly via construction of data centers, and indirectly via valuation effects on consumption. And then the multipliers kick in. If not for sky-high optimism about AI, the economy would now be in recession with high probability. This sky-high optimism has pushed stocks up to remarkably high valuation ratios rivaling those of the very end of the 1990s:
And this in spite of long-term bond rates that are quite low in historical perspective—inconsistent with any sort of strong expected future productivity boom that might generate income to validate such high valuation ratios. Low bond rates even when there is a furious AI bubble-driven data-center capital investment boom—they are certainly registering the presence of a global savings glut, if not of full-blown secular stagnation:
As best as I can see, this sky-high optimism about stocks is not warranted.
And here is the very sharp Gillian Tett is right now sounding the alarm about the shift in AI data-center funding to debt and to circular vendor financing. When the people who are giving you the money do not trust that your speculative enterprise, that is a bad sign. And when you can only get customers by lending them the money to buy your products—that is a very bad sign too. Gillian Tett:
Gillian Tett: AI has a cargo cult problem <https://www.ft.com/content/f2025ac7-a71f-464f-a3a6-1e39c98612c7>: ‘Spending vast sums and inflating an investment bubble is no guarantee of unleashing technological magic: Is it just a Ponzi scheme? That is the question that currently haunts American tech — and wider markets — as the valuation of artificial intelligence-linked groups soars to evermore eye-popping heights…. Ten lossmaking AI start-ups — such as OpenAI, Anthropic and Elon Musk’s xAI — now command a collective valuation of close to $1tn, while venture capital has poured $161bn into AI overall this year…. Fw of these entities expect to turn a profit anytime soon — and these valuations are being boosted by variants of cross-cutting vendor financing, like recent deals between OpenAI, Nvidia, Oracle, AMD and Broadcom… [in] a pattern of circular flows that echo some of the hairball of interconnections that emerged between banks and insurance companies via credit derivatives before 2008… [which] resulted in unseen concentrations of risk — and subsequent contagion when the bubble burst…. And… even tech luminaries, like Jeff Bezos, admit there is excessive exuberance…
But, right now, it is a bad sign only for those invested in the technology’s long-term superprofitability and in making money by building and running data centers. For while it is to some degree a Ponzi scheme, it is not “just” a Ponzi scheme. And that makes a big difference: that means that it will probably go on for quite a while, and that when there is the ultimate shakeout the tech world will look very different and very weird because of all the money poured into data centers, software development, and business-model experimentation.
Back up. These days I like to say that the AI bubble is eight things. It is:
one part: reasonable expectation of providing and financially capturing true end-user value,
two parts: millennarian religious hype on the part of those hoping for the Rapture of the Nerds,
three parts perhaps-reasonable expectation of improved advertising targeting for user good and user ill,
four parts: grifters seeking easy investor marks moving over from crypto,
five parts: platform near-monopolists fearing the loss of their profit flows to a Christensenian disruption from the Next New Big Thing,
six parts:
speculative possible enormous increases in utilizer surplus from the ability to add natural-language interfaces to every interaction with structured and unstructured databases,
or decreases in human flourishing as natural-language interfaces serve as a stalking-horse for yet another round of hacking users’ attention, and not for their benefit,
but in either case not a likely source of Google-Facebook-Apple-style platform monopoly profits,
seven parts: the downstream consequences for human society and culture flowing from enormous increases in human data-analysis capabilities driven by the unbelievably large scope of huge data, enormous dimension, highly flexible-function classification tools coming online now and soon, and last:
eight parts: the downstream consequences for human society and culture flowing from MAMLM-mediated—Modern Advanced Machine-Learning Model-mediated—human interaction with the infosphere.
Of these, only (1) and (3) are likely sources of superprofit for investors.
(6), (7), and (8) are transformative, potentially, for human society and culture, but not likely sources of superprofits from investors in providing AI-services or AI-support right now. (5) is a defensive move: not an attempt to boost the profits of platform monopolists, but to spend a share of those profits—a large and growing share—defending them against Clayton Christensenian disruption. (2) and (4) are culturally important, and are driving much of investor interest on the belief that with so much excitement about this pile of manure there must be a pony in there somewhere. But focusing on them is not likely to lead to good investment decisions.
Yes, the grifters moving over from crypto are definitely trying to run a Ponzi scheme: if the person trying to sell you something has just spent a decade selling BitCoin, DogeCoin, and Web3 use cases coming real soon now, you should probably block them and add them to your spam list.
And anyone saying either of these two things: (i) soon everyone will be under threat from a malevolent digital god that will soon control all of our minds through flattery, misdirection, threats, and sexual seduction, the negative millennarians; or (ii) they are on the cusp of building a benevolent digital god. Shake your head and walk away. And if they then turn to “give me money! lots of money!”—well, then run.
oth groups are mostly conning themselves. But do not believe any forecasts or accept any unsecured debt obligations from anyone talking about ASI—Artificial Super-Intelligence—coming within a decade.
And if whatever project is pitched to you requires that the enterprise make a lot of money by using some form of natural-language interface to eat away at and grab a share of Google’s, Facebook’s, Amazon’s, or Apple’s platform-monopoly profits, run away: Google, Facebook, Amazon, and Apple will in the end put five times as many programmers on-task as whatever you have invested in, and give away their product for free to keep you from being successful. Think: Microsoft vs. Netscape in the 1990s. Oracle, Microsoft, and OpenAI may make money at a reasonable pace by selling data-center and chatbot-interface services; firms like CoreWeave, Nebius, Lambda, Equinix, Aligned, Digital Realty may make substantial amounts of money too; as will AMD, BroadCom, TSMC, ASML and most of all NVIDIA (but those are all very richly valued by now even for the likely future profit flows).
But the market disagrees. So we do read, day by day, about things like:
Ruchir Sharma: America is now one big bet on AI <https://www.ft.com/content/6cc87bd9-cb2f-4f82-99c5-c38748986a2e>: ‘Large companies and investors… are increasingly confident that artificial intelligence is such a big force, it can counter all the challenges. Lately, this optimism has become a self-fulfilling prophecy. The hundreds of billions of dollars companies are investing in AI now account for an astonishing 40 per cent share of US GDP growth this year…. AI companies have accounted for 80 per cent of the gains in US stocks so far in 2025. That is helping to fund and drive US growth, as the AI-driven stock market draws in money from all over the world, and feeds a boom in consumer spending by the rich….
Without all the excitement around AI, the US economy might be stalling out, given the multiple threats.… [The] immigration boom-bust cycle… alone will reduce America’s growth potential by more than a fifth…. Government deficits and debt are increasing faster in the US than in other developed markets…. AI is regarded as a magic fix… [because] it is expected to deliver a significant boost to productivity growth…. The possibility of a productivity miracle to come has cemented the faith of investors…. Foreigners poured a record $290bn into US stocks in the second quarter and now own about 30 per cent of the market…. AI [had] better deliver for the US, or its economy and markets will lose the one leg they are now standing on…
If anything, this is understated:
It is not that the U.S. economy “might be stalling out”.
It is that the U.S. economy would be stalling out, or worse.
We will have a truly ungodly huge amount of datacenters and GPUs and (largely natural gas) power plants when this build-out is done. For example, only the first of many is StarGate 1 in Abilene, Texas: 500,000 square feet and 200 MW per building; 8 buildings planned; a campus roughly 1 square mile.
These concentrations of computer power will be useful for something.
Why are we here? Well, I gave eight reasons above, of which only one, and of that the smallest, was the idea that those undertaking the programming and data-center building-out are doing so as a result of their reasonable expectation of providing and financially capturing true end-user value.
But there is also a ninth reason. The ninth reason is that, right now, the bear case is simply not being made by anyone outside the extended AI-knowledgeable community.
I think the extent to which that case is not being made is best grasped by looking at Matt Yglesias here:
Matt Yglesias: The AI boom is propping up the whole economy: It’s a fragile basis for growth, whether or not it’s a “bubble.”… When we were pitching the idea that became Vox… the people who were interested in the pitch were interested in Ezra Klein and Melissa Bell and Matt Yglesias. And that was not a crazy calculation…. And the other thing… is that, at the time, investors were enthusiastic about ad-supported digital media… [on the] thesis (wrong, as it turned out) that you could make a lot of money…. That’s where it gets hard to judge these things. Investing in start-ups… [as a] business… depends less on having a very high batting average than on the idea that home runs will be extremely lucrative…. The thesis lives or dies primarily on the question of how big the hits will be. And if I could tell you that, I wouldn’t be writing a subscription newsletter for a living…. The leading AI labs are genuinely seeing rapid revenue growth. I don’t think “the trajectory of model capabilities improvement and associated revenue will continue on its current path indefinitely” is the most sophisticated investment thesis in the world. But it’s also not unmoored from reality. It’s a plausible but possibly wrong guess about the future state of the world — i.e., a pretty normal investment hypothesis…. The nice thing about financial markets is that the people investing in them have very strong incentives to at least try to avoid wishful thinking and motivated reasoning and instead focus on cold hard cash…
And here Matt has, I think, served his readership badly.
He should say: The question of how big the hits will be and whether you or anyone should invest hinges on the interaction of (i) the technology and its future development, (ii) the industrial market structure in that it has to be conducive to both the commodification of all of the pieces of the value chain complementary to your enterprise and durable moat-construction for your piece, and (iii) the demand which needs to be potentially sky-high.
He should also say: If someone—or the market as a whole—brings you what they claim is such an opportunity, remember that there is a famous passage in the American musical “Guys & Dolls”:
Sky: Nathan, let me tell you a story.
Nathan: Have we got a bet?
Sky: Nathan, let me tell you a story. On the day I left home to make my way in the world, my daddy took me to one side.
“Son,” my daddy says to me, “I am sorry I am not able to bankroll you to a large start, but not having the necessary lettuce to get you rolling, instead, I’m going to stake you to some very valuable advice.
“One of these days, a guy is going to show you a brand-new deck of cards on which the seal is not yet broken. Then this guy is going to offer to bet you
that he can make the jack of spades jump out of this brand-new deck of cards and squirt cider in your ear.“But, son, you do not accept this bet because, as sure as you stand there, you’re going to wind up with an ear full of cider…”
He should then also say: I do not believe I can judge the chances of a big hit—the only people who can are the people who know enough about industrial structure and user demand to get by but know a huge amount about the technology, the people who know enough about the technology and about potential user demand to get by but know a huge amount about industrial structure, and the people who know enough about the technology and industrial structure to get by and know a huge amount about potential user demand.
And he should penultimately say: If the person offering you this deal is not in one of those three groups—and the market as a whole right now definitely isn’t—what the hell are you doing listening to their representations that this is one of those cases in which you should invest anyway? And if the person offering you this deal is in one of those three groups—and so knows whether this is actually a favorable-odds bet or not—what is their motive for offering you a piece of the action rather than keeping it all for themselves and their friends?
And so the ultimate punchline should be quite different than what Matt says. Matt says that it may well not be a bubble—that investments in AI programming and the AI build-out right now may well make sense—because people in financial markets “have very strong incentives to at least try to avoid wishful thinking and motivated reasoning”, and hence Democrats should think very hard about “how to win in 2028 if unemployment stays low and growth remains robust”.
But look at the eight things driving this—reasonable profit expectations, another ride of the crypto-grifters, millennarian religious hype, advertising targeting, platform-monopoly defense, huge but nearly impossible to turn into profit shifts in human behavior, the wild card of enormous new analytical capabilities, and the downstream consequences of a large shift in the modality of human communication.
Look at all that, and the obvious conclusion is that the fact that this is a bubble does not mean any sort of economic bill in terms of high unemployment and falling production is going to come due.
Trees don’t grow to the sky. But bubbles that are far more complicated than simple Ponzi schemes do not collapse on any schedule. Especially if the Ponzi-scheme element is only a very small part of the total econo-techno-cultural-socio story.
"We will have a truly ungodly huge amount of datacenters and GPUs and (largely natural gas) power plants when this build-out is done. For example, only the first of many is StarGate 1 in Abilene, Texas: 500,000 square feet and 200 MW per building; 8 buildings planned; a campus roughly 1 square mile.
These concentrations of computer power will be useful for something."
What's the rate of deprecation on a data center? GPUs burn out pretty fast right?
Railroads move goods, radios move audio, the Internet moves data, houses provide shelter, and AI ??? We're forcing AI to fit into social media and search engines because these have high profit margins for a few investors.
What if the prevalent use of AI isn't to attract eyeballs or replace white collar labor, but to automate manual labor? That's a boring abstraction in Silicon Valley. Imagine that China creates dexterous, sensing mechanical hands that can quickly assemble everything from I-phones to T-shirts. That wouldn't use giant data centers, the latest Internet data, LLM's, or produce video. I'm not predicting that, but I am suggesting that smaller, specialized models built for specific applications may be more revolutionary than giant solutions investments in search of a problem.