Michael Burry has seen a financial bubble before.
Now he thinks he is watching another one form.
But his latest warning about artificial intelligence is more interesting than the usual argument that AI stocks are simply "overvalued."
The concern is about how the AI boom is being financed.
According to an Ares analysis cited by Burry, roughly $573 billion across 26 disclosed AI-related financings over the previous 12 months has created a web connecting some of the industry's biggest companies through debt, leases, guarantees, chip purchases and customer contracts.
And that creates a potentially uncomfortable question:
How much of today's AI growth represents genuine end-user demand—and how much is being reinforced by the companies building the AI ecosystem financing one another?
Follow the money
Consider the basic structure.
Nvidia supplies the chips.
Cloud companies buy the chips.
AI companies rent the computing capacity.
Cloud providers build data centres.
Investors finance the data centres.
AI companies sign enormous future contracts for computing.
Those contracts can help support financing.
And some technology companies provide guarantees or other forms of support that make the financing possible.
The result is an ecosystem in which the same companies can simultaneously be:
supplier → customer → investor → creditor → guarantor.
That doesn't automatically mean fraud.
It doesn't even necessarily mean the transactions are bad.
But it creates interdependence.
And interdependence becomes dangerous when everybody's business model depends on the same assumption.
The assumption is simple
AI spending keeps growing.
Microsoft keeps building.
Meta keeps building.
Amazon keeps building.
Oracle keeps building.
OpenAI and Anthropic keep consuming enormous amounts of computing capacity.
Nvidia keeps selling GPUs.
Broadcom keeps supplying critical infrastructure.
Investors keep financing data centres.
And everyone assumes the next customer will arrive.
Then the next one.
Then the next one.
This is why Burry says the AI bubble is effectively on a clock.
He has previously argued that increasing reliance on debt financing means the cost of capital eventually becomes a constraint, with 2028 as his base case for when the financing structure becomes particularly vulnerable.
His newest warning is therefore less about whether AI works.
It is about whether the financial system supporting the AI buildout can continue expanding faster than the underlying economics justify.
The telecom comparison
Burry compares the current structure with the late-1990s telecom boom.
That comparison isn't completely crazy.
During the telecom boom, companies spent enormous amounts building fibre-optic networks based on expectations of explosive future internet traffic.
The infrastructure was real.
The technology was real.
The demand was real.
But investors massively underestimated how much capacity would eventually exist relative to what customers were willing to pay for.
When expectations collapsed, the financial structure supporting the buildout collapsed with them.
The important lesson wasn't that fibre was useless.
It was that a transformative technology can be real and its financial bubble can still be real.
AI can be both.
And there is already evidence of financial stress
This isn't merely Burry inventing a hypothetical scenario.
The Bank for International Settlements has separately warned about circular financing patterns in AI, identifying arrangements where firms receive financing and business from the same counterparties, suppliers finance customers to support demand, or customers finance suppliers to secure critical inputs.
The Federal Reserve has also examined the financial risks.
The Chicago Fed estimated that large-bank commitments to AI-adjacent industries had reached roughly $450 billion by late 2025, while warning that a severe deterioration in AI investment could produce knock-on effects across software, data centres, energy and semiconductor companies.
And S&P Global expects combined hyperscaler capital expenditure to exceed $1.3 trillion by 2027, with debt, leases, guarantees and other financing structures playing an increasingly important role.
That is an extraordinary amount of capital being committed to one technological investment cycle.
Nvidia is turning compute into an asset class
This is where the story gets particularly interesting.
Nvidia is not merely selling GPUs anymore.
It is helping construct financing mechanisms around AI infrastructure.
In August, Jensen Huang announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR designed to mobilize more than $500 billion of third-party capital for AI infrastructure over time.
Nvidia explicitly describes AI compute as becoming an investable infrastructure asset.
There is nothing inherently wrong with that.
Airports, power plants, fibre networks and data centres are all financed as infrastructure.
But AI introduces an unusual problem:
technology depreciates incredibly quickly.
A power plant doesn't become obsolete because a new version of electricity was invented.
A GPU can.
That is why Wall Street lenders have been questioning how long expensive AI chips should be treated as economically useful assets.
Reuters reported that lenders were reluctant to value GPUs using Nvidia's suggested useful life of up to a decade and were instead looking at much shorter depreciation schedules, often around three to four years.
That matters enormously.
If the hardware is worth less than expected before the debt is repaid, somebody absorbs the difference.
But Burry could still be wrong
This is the part missing from many viral versions of his argument.
AI demand isn't imaginary.
Nvidia has enormous revenue.
Microsoft, Amazon, Meta and other hyperscalers generate enormous cash flows.
AI is already being used commercially.
And the infrastructure being built today could support applications that haven't even been invented yet.
The Financial Times reported this week that analysts expect S&P 500 earnings to rise about 27% year-over-year in the September quarter, with AI-heavy companies among the major contributors. Hyperscaler capital spending is expected to rise dramatically as well.
So this isn't 1999 all over again.
The companies driving today's boom are generally much larger, more profitable and better capitalized than the speculative internet startups that dominated the dot-com bubble.
That's the strongest argument against Burry.
The real danger isn't AI failing
It is AI economics failing to grow fast enough to justify the infrastructure being financed today.
Imagine AI revenue keeps growing—but only 20% instead of 100%.
Imagine enterprises discover that some AI workloads aren't worth the cost.
Imagine model prices continue collapsing.
Imagine newer chips make yesterday's expensive GPUs less valuable.
Imagine AI companies renegotiate enormous computing commitments.
Imagine investors demand higher returns.
Suddenly, the problem isn't that AI stopped working.
The problem is that the financial assumptions surrounding AI stop working.
And because the ecosystem is interconnected, one company's slowdown can become another company's problem.
A weaker AI customer means less demand for cloud capacity.
Less cloud demand means pressure on data-centre economics.
That affects infrastructure financing.
That affects hardware demand.
That affects chip suppliers.
And investors then reassess the valuations of the companies sitting on top of the entire system.
That's the loop Burry is worried about.
The question investors should be asking
Not:
“Is AI real?”
It obviously is.
Not:
“Will AI transform the economy?”
It probably will.
The more important question is:
“How much future AI growth has already been priced into today's infrastructure, contracts, valuations and debt?”
Those are very different questions.
A technology can change the world while investors lose money betting on it at the wrong price.
Railways changed the world.
Electricity changed the world.
The internet changed the world.
Telecommunications changed the world.
None of that prevented enormous amounts of capital from being destroyed during the financial bubbles surrounding those technologies.
Burry's warning is therefore worth listening to—but not because he predicts a crash
Michael Burry has become famous for predicting financial disasters.
That also makes him vulnerable to becoming a character in his own narrative.
A famous bear saying "the crash is coming" is not evidence that a crash is coming.
The more useful part of his argument is buried underneath the headline.
Look at the financing.
Look at who is lending.
Look at who is guaranteeing.
Look at who is buying.
Look at who is supplying.
Look at who is signing future contracts.
Then ask how much of the ecosystem depends on everybody continuing to spend.
Because the greatest vulnerability of a financial bubble isn't necessarily that the underlying technology is fake.
Sometimes the technology is revolutionary.
The problem is that the financial system gets ahead of the technology.
And if AI spending eventually slows, the world may discover that some of the industry's spectacular growth wasn't simply generated by AI demand.
It was also generated by money chasing money around the same AI ecosystem.
That is the warning investors should take seriously.
Not because Burry has a crystal ball.
But because a boom financed by interconnected promises becomes fragile when everyone needs everyone else to keep spending.
Comments
Post a Comment