Wednesday, August 19th, 2026
Bill Bonner, from Poitou, France
The pin approaches. It is a patient pin. It will take its time, but not forever. In all of the bubbles we know of — from tulips to railroads to suburban mortgages — it has never missed its mark.
But when we turn our eyes to the bubble itself, we are shocked. This is no ordinary bubble. This is a SuperBubble...bigger...more flamboyant...and more charged with gas than any that came before it. When the pin gets close, watch out.
But first...a ray of sunshine on a gloomy day. Bloomberg:
Parents are hiring $6,500 sorority consultants
Worried about people being out of work because of AI? Well, you can stop worrying. People who will pay a consultant to help a daughter choose a sorority will never lack for ways to take on new hires. But the transition could be rough. And when pin and bubble finally meet, the poor girl can go back into the dorm.
Between the election of Obama and today, the feds have added about $30 trillion of debt — three quarters of the total. So much debt has been added in the last 18 years that a real correction can’t happen without massive bankruptcy, chaos and inflation. That $30 trillion (we are not including private sector credit) was ‘ghost demand’ that came out of nowhere. Nobody earned it. Nobody saved it. But it added to sales, profits...capital values...and the SuperBubble.
The total of US debt in 2008 was only $10 trillion. Now it is approaching $40 trillion — four times as much. And the cost of servicing the debt this year is $1.3 trillion. In 2008, the interest cost was still less than $200 billion...only a fraction of today’s bill.
AI has an important role to play in the SuperBubble too. Just as Alan Greenspan anticipated a big boost to output from the internet (giving him cover to increase the money supply), today Kevin Warsh believes AI will do the job:
“AI will be a significant disinflationary force, increasing productivity and bolstering American competitiveness.”
So far, there is no sign of it. Like the internet itself, AI is focused on the ‘paper industry’ of ‘knowledge workers’ — ideas, analyses, contracts, reports, plans, programs, entertainments. The ‘paper’ is a cost...not an asset or revenue. And if the lessons of the internet apply, the machine will produce more paper, making business more expensive, slower...and requiring three more meetings.
Last week, we looked at how incestuous the AI complex had become. A dollar of speculation may show up as income in several places — boosting capital values in all of them.
But where did the dollar of investment come from? Where was it earned? When was it saved?
Maybe nowhere and never. Like federal debt, it comes out of nowhere...from the credit market or from the bubble itself. In the case of Oracle, for example, the company is putting more than 100% of its earnings into its AI bet. Other speculators are drawing on their bubble gains elsewhere to keep the bubble expanding. The Nasdaq has doubled over the last four years; there’s plenty of wealth to draw on. But it’s bubble wealth, not real wealth. Actual output rises only half as fast as the stock market. Myrmikan Research:
The hyperscalers plus Nvidia plus the three U.S.-based Nvidia suppliers comprise 29.7% of the S&P500—eighty million working Americans every two weeks have part of their paychecks shunted into retirement accounts that buy that index. The inflated market caps provide the collateral that backs over $600 billion in margin loans, used either to boost consumer spending or to fund more stock speculation. An additional $152 billion of consumer debt has been borrowed by life insurance policy holders against their policies, that is, debt borrowed against the debt owned by the insurance companies, increasingly comprised of illiquid private credit held offshore.
In what might be a peak for the whole Bubble Epoch, Google sold a 100-year bond in February. You may wonder why a $4 trillion company...with earnings of $400 billion...would want to borrow at all. Or why sane investors would want to lend money to a company that doesn’t need the money and probably won’t be in business when its bonds finally mature? And what about the dollar; will it be just a collector’s item? But that is just the nature of a SuperBubble. You may already be hoist to dizzy heights, but there is nothing that says you can’t go higher.
Except history. And math. And newer tech. And competition. And physics. The gap between fetching up info, as Google does, and analyzing it, as Claude does, is wider than most people think. It turns out that it takes much more energy to ‘think’ than to fetch. That’s why there are so many ‘data centers’ popping up in cornfields throughout the nation. But it might turn out that the real return is much lower than expected. Myrmikan Research:
According to an April 2026 academic paper: “agentic tasks are uniquely expensive, consuming 1000X more tokens than code reasoning and code chat…” Even worse for businesses: “Task difficulty rated by human experts only weakly aligns with actual token costs, revealing a fundamental gap between human-perceived complexity and the computational effort agents actually expend…Frontier models fail to accurately predict their own token usage….and systematically underestimate real token costs.”
And here’s more good news, recent surveys hint that humans may already be making a comeback. The cost of AI is so high, in some cases, that humans can be cheaper. More effective. And they’ll sit down and have a cup of coffee with you. What’s more, China is developing its own AI...and, as in so many other things, seems to be undercutting US prices. In the normal course of events, it could turn out that the AI industry is smaller and less profitable than we imagined...and that the Chinese own it. Two Chinese companies — Deepseek and Moonshot — are already on their way, according to press reports.
So, how big is the SuperBubble? The ‘hyperscalers’ — Microsoft, Amazon, Google, et al — are worth about $16 trillion. The chipmakers — Nvidia, et al — are worth about $10 trillion. And the AI companies themselves — Anthropic, OpenAI, xAI et al — are worth about $4 trillion more. That’s $30 trillion altogether. Add the feds’ unpayable debt, another $30 trillion, and you have $60 trillion of hot air.
A trillion here. A trillion there. Pretty soon you’re talking about one helluva bubble.
Regards,
Bill Bonner
WELCOME TO THE ARTIFICIAL INTELLIGENCE REVOLUTION!
I’m fascinated by the 100-year bond. A hundred years, that’s confidence.
You lend a corporation money in 2026 and they promise:
“Don’t worry, we’ll settle up in 2126.”
Oh, excellent, I’ll put it in my calendar.
Right between: “Colonize Neptune” and “Annual prostate exam.”
Who buys a 100-year bond?
You can’t even get a contractor to guarantee a WATER HEATER for twelve years. But apparently investors look at a technology company and say:
“Surely nothing important will change over the next century.”
Really?
One hundred years ago, people were listening to radio programs beside a stove that could kill them. Now we’re asking a computer to write erotic fan fiction about the toaster. Things change. Imagine buying a 100-year technology bond in 1926.
“Gentlemen, I’ve invested the family fortune in the future!”
“What company?”
“Telegraph poles.”
“How long is the bond?”
“A century.”
“Wonderful. Your great-grandchildren will receive eleven dollars and a museum exhibit.”
And now we have AI companies valued at astronomical numbers. Wonderful businesses. Some make money. Some expect to make money. Some are investing gigantic sums so that someday they can make money. And some apparently have achieved the most advanced financial technology ever invented:
MAKING THE WORD “SOMEDAY” WORTH $200 BILLION.
That’s incredible. I tried that with my mortgage. The bank wasn’t impressed.
“Sir, when will you make the payment?”
“Someday.” Apparently, I needed a pitch deck.
Computers will replace humans!
EVERYBODY PANIC! The machine works twenty-four hours a day!
Never sleeps! Never complains! Never asks for health insurance! Never takes vacation! Fantastic!
How much does it cost?
DON’T ASK THAT.
How much electricity?
NEXT QUESTION.
How many data centers?
LOOK OVER THERE! PRODUCTIVITY!
Suddenly Iowa has a building consuming enough electricity to open a portal to another dimension.
Farmer walks outside:
“Margaret, why is the corn glowing?”
“Claude is making a PowerPoint.”
And apparently some AI tasks consume enormous quantities of computational resources. Beautiful. We spent decades complaining:
“Damn employees! Coffee breaks! Lunch breaks! Health insurance!”
So we replaced Steve from Accounting with a machine. Steve cost $72,000 a year. The machine requires:
TWO ACRES OF NVIDIA CHIPS, A SMALL NUCLEAR REACTOR AND THE COLORADO RIVER.
Management: “LOOK AT THE SAVINGS!”
Then somebody discovers humans might occasionally be cheaper.
THE HUMAN RETURNS!
Corporate America:
“Wait…the biological version runs on sandwiches?” YES!
“And coffee?” YES!
“Can it attend meetings? Unfortunately.
“Can it produce reports?” Yes.
“Can it explain why we’re having the meeting?”
NO TECHNOLOGY HAS ACHIEVED THAT.
And suddenly the human employee becomes revolutionary technology.
Introducing: HUMAN 2.0, Runs on approximately 2,000 calories per day! Portable! Self-cooling! Wireless! Can recognize sarcasm! Requires no data center! Occasionally smells strange!
And after work…
IT TURNS OFF.
Silicon Valley:
“My God.”
If the NPCs would acquire their knowledge the old fashioned way by reading physical books instead of issuing their drive-by queries to "AI", collective IQ would increase and the AI industrial complex would go the way of the internet - a novelty at first but now just a moderately useful tool for information sharing; and for the masses it's a useless distraction and huge waste of time.