AI can change the world, revenues can keep exploding, demand can remain strong—and investors can still lose money because the price paid and the amount of capital invested matter.
Follow Michael Burry here:
Artificial intelligence may turn out to be one of the most important technological developments of our lifetime.
AI revenues could continue growing.
Demand for computing power could continue exploding.
Microsoft, Amazon, Google, Meta and Oracle could report record AI-related revenue.
Nvidia could continue selling enormous numbers of chips.
Companies could continue telling investors that they cannot build data centers fast enough to satisfy demand.
And Michael Burry could still be right.
That sounds contradictory, but it really isn’t.
Burry’s warning is not simply that AI is a fad or that nobody will use it.
His argument is much more interesting for investors:
A great technology does not automatically make every investment in that technology a great investment.
And perhaps more importantly:
The better today’s AI demand looks, the more money companies are encouraged to spend chasing it. At some point, that enormous spending can itself become the problem.
That distinction is the heart of Burry’s latest research and something every investor should understand.
The Most Important Idea: Growth and Investment Returns Are Not the Same Thing
Let’s start with a simple example.
Imagine there is a city with 100,000 pickleball players and not nearly enough courts.
Every facility is packed.
Court reservations sell out immediately.
Membership prices rise.
Existing facilities report record revenues.
Every piece of evidence says exactly the same thing:
Pickleball demand is booming.
Investors see the opportunity.
One developer builds 20 courts.
Another announces 30.
A private-equity firm finances another 50.
Existing clubs expand.
Real-estate developers convert warehouses.
Banks happily lend money because every existing facility is full.
Within several years, instead of 100 courts, the city has 1,000.
Pickleball didn’t fail.
People didn’t stop playing.
There might even be more players than ever before.
But something important changed.
There are now too many courts relative to demand.
Court prices fall.
Facilities compete for members.
Margins shrink.
Some highly leveraged operators cannot service their debt.
The facilities that were built assuming today’s extraordinary economics would continue indefinitely suddenly don’t produce the expected returns.
The sport succeeded.
The investment boom failed.
That is a useful way to understand Michael Burry’s concern about AI.
Burry Is Watching the Capital Cycle
Burry calls this the capital cycle.
It is a pattern that has occurred repeatedly throughout financial history.
It generally looks something like this:
1. A new opportunity appears.
Demand exceeds supply.
2. The early companies make extraordinary profits.
Investors notice.
3. Capital floods into the industry.
Companies borrow money, raise capital and dramatically increase spending.
4. Everyone expands at approximately the same time.
The shortage encourages even more investment.
5. Eventually all that new capacity arrives.
Supply begins catching demand.
6. Pricing power weakens.
Returns on all the new investment decline.
7. Investors finally realize that too much capital was invested.
The stocks can fall well before the underlying industry actually stops growing.
Burry believes we are already deep into this process with AI.
He calculates that S&P 500 net investment relative to the size of the U.S. economy is around levels exceeded in recent decades only around the technology bubble. He specifically points out that during the dot-com era, investment continued climbing after the Nasdaq had already peaked.
That historical point matters.
The stock market does not necessarily wait until the factories are empty or data centers are losing money.
Markets anticipate.
The stocks can recognize the problem before the capital spending itself peaks.
The Numbers Behind Burry’s Concern Are Enormous
Burry has been examining five companies at the center of the AI infrastructure buildout:
Microsoft
Amazon
Alphabet/Google
Meta
Oracle
These are often called the AI “hyperscalers.”
Burry estimates that their combined future purchase commitments, leases, guarantees of third-party debt and various potential off-balance-sheet obligations amount to roughly:
$3 TRILLION
And much of that money has not been spent yet.
It represents commitments that can keep the AI infrastructure boom running for years.
This is where Burry’s argument becomes especially interesting.
The risk isn’t simply how much has already been spent.
It is how much has already been promised for the future.
Today’s AI Numbers Can Look Fantastic While Tomorrow’s Economics Get Worse
Imagine Microsoft reports:
AI revenue +70%.
Amazon reports:
AWS AI demand +80%.
Google reports:
Cloud backlog +100%.
Oracle reports:
Record remaining performance obligations.
Nvidia reports:
Another record quarter.
Most investors naturally interpret this as:
“AI is working. Buy AI.”
Burry asks a different question:
What are companies spending today to produce all of that future revenue?
And then:
What return will they ultimately earn on that spending?
Those are very different questions.
A company can generate another $10 billion of revenue and still destroy shareholder value if producing that revenue requires $15 billion or $20 billion of investment that earns an inadequate return.
Revenue growth by itself does not tell us whether the investment was good.
The Irony: The Better AI Looks Today, the Greater the Risk of Overbuilding
This is probably the most counterintuitive part of Burry’s warning.
Suppose AI demand suddenly weakened today.
Companies would probably cancel projects.
Data centers wouldn’t get built.
Chip orders would fall.
Capital spending would decline.
Ironically, that would prevent some future oversupply.
But that’s not what’s happening.
Demand is exceptionally strong.
Microsoft itself described the current supply-demand environment as an unusually extreme situation in which demand exceeds available supply.
That encourages everybody to build more.
And that’s precisely how capital cycles become excessive.
Shortages create high prices.
High prices create high profits.
High profits attract capital.
Capital creates new supply.
Eventually new supply destroys the shortage that created those extraordinary profits.
In other words:
Today’s fantastic economics encourage the investment that can undermine tomorrow’s fantastic economics.
Microsoft: What If the Assets Don’t Last as Long as the Accounting Assumes?
Microsoft provides an interesting example.
According to Burry’s analysis, Microsoft’s traditional bond debt declined from approximately:
$43.2 billion to $40.3 billion.
That sounds conservative.
But its finance-lease liabilities increased from approximately:
$46.2 billion to $66.6 billion.
Burry also focuses heavily on Microsoft’s assumptions regarding how long its data-center assets will remain useful.
Microsoft extended certain assumed useful lives from approximately 15 years to as much as 25 years.
Burry argues that longer assumed lives can influence depreciation and lease classification and therefore affect how the economics appear in the financial statements.
Forget the accounting terminology for a moment.
The basic investor question is very simple:
How long is AI equipment really economically valuable?
The building might physically last 25 years.
The concrete will still be there.
The power connections will still work.
The cooling infrastructure may still function.
But what about the expensive technology inside it?
AI chips can become technologically outdated remarkably quickly.
So an investor needs to distinguish between:
physical life
and
economic life.
A computer may physically operate for 15 years.
That doesn’t mean you would want to run a competitive AI business using a 15-year-old computer.
Microsoft Has Another Issue: Enormous Future Commitments
Burry also highlights approximately $329 billion of signed but not-yet-started leases at Microsoft.
Think about what that means.
Microsoft hasn’t simply said:
“If AI demand remains strong, maybe we’ll build more.”
A meaningful amount of future infrastructure is already contracted.
That’s important because management often says that if AI demand slows, capital spending can be reduced.
That may be true for future decisions.
But it’s harder to change decisions you’ve already contractually committed to.
Amazon: Follow the Cash, Not Just the Revenue
Amazon provides another example of why Burry wants investors looking beyond headline growth.
According to his analysis, Amazon’s combined uncommenced leases and purchase commitments have reached approximately:
$267 billion
And they increased roughly 81% in nine months.
Meanwhile, Amazon’s backlog increased from approximately:
$200 billion to $496 billion
in only nine months.
Sounds fantastic.
But something else happened.
The weighted average duration of that backlog increased from approximately:
3.8 years to 6.4 years.
That distinction matters enormously.
Imagine two companies both tell you:
“We have $100 billion of future orders.”
Company A will receive nearly all that revenue over the next 18 months.
Company B expects much of it six or seven years from now.
Those aren’t economically identical.
The further into the future revenue sits, the greater the uncertainty.
Will the customer still need the capacity?
Will the customer remain financially healthy?
Will today’s prices still exist?
Will competitors have cheaper technology?
Will computing become dramatically more efficient?
Will new chips make existing infrastructure obsolete?
The number investors should care about isn’t simply:
How large is the backlog?
It should also be:
How good is the backlog?
Amazon’s Financing Is Changing Too
Burry highlights another striking comparison.
In the twelve months through June 2025, Amazon reportedly raised approximately:
$746 million of long-term debt.
In the comparable twelve months through June 2026:
$81.9 billion.
At the same time, Burry calculates free cash flow at roughly negative $8 billion.
This doesn’t mean Amazon is financially distressed.
That’s not the argument.
Amazon is one of the largest companies in the world.
The point is the direction and magnitude of change.
The AI investment race is becoming extremely capital intensive.
Meta: Debt Doesn’t Always Look Like Debt
One of the recurring themes in Burry’s work is that investors need to look beyond the traditional debt line on a balance sheet.
Meta illustrates why.
Burry calculates roughly $700 billion of uncommenced leases and purchase obligations, including certain commitments that don’t appear as conventional balance-sheet debt.
Some infrastructure can be financed by third parties.
Someone else borrows the money.
Someone else builds the facility.
But Meta may provide guarantees or contractual commitments that make the financing possible.
Imagine I want a $100 million building.
Instead of borrowing $100 million myself, I tell another company:
“You borrow the money and build it. I’ll lease it from you for many years, and I’ll provide certain guarantees.”
Technically, my conventional bank debt might still look low.
Economically, however, I’ve made a substantial long-term commitment.
Burry wants investors to understand that distinction.
Google: Almost $900 Billion of Potential Exposure
Alphabet may be the most striking example.
Burry estimates Alphabet has approximately:
$707 billion of supply-chain purchase obligations
and total commitments and potential off-balance-sheet exposures approaching:
$900 billion.
Again, Google is enormously profitable.
That’s not the point.
The question is:
How much future profitability is already being committed today to win the AI race?
And Burry identifies another concern.
Circular Financing: When Everyone Is Funding Everyone Else
This deserves special attention.
The AI ecosystem is increasingly interconnected.
Big technology companies invest in AI companies.
AI companies raise enormous amounts of capital.
Those AI companies use the money to purchase computing services from the same ecosystem.
Cloud companies use the expected demand to justify building more infrastructure.
Lenders finance the infrastructure because large technology companies provide contracts or guarantees.
The infrastructure providers then buy enormous quantities of chips.
Chip companies invest throughout the AI ecosystem.
And around the circle we go.
For example, Burry discusses Google’s relationship with Anthropic.
Alphabet owns a stake in Anthropic and is investing substantial additional capital. Meanwhile, financing has been raised for Google TPU chips intended to provide computing capacity to Anthropic. Google can then include relevant contracts in its cloud backlog.
This doesn’t mean the revenue is fake.
That’s an important distinction.
But Burry’s question is:
How much of the extraordinary growth is being financed by capital flowing around the same ecosystem?
That matters because circular systems work beautifully while financing remains abundant.
They become much more vulnerable when capital becomes expensive or investors stop providing additional money.
Recent reporting also shows that bond investors have become more selective about AI-related debt as the industry’s borrowing requirements grow. (Reuters)
Oracle: $664 Billion of Backlog Sounds Amazing — Until You Ask When It Arrives
Oracle’s reported backlog is approximately:
$664 billion.
That’s an extraordinary number.
But Burry digs deeper.
Approximately:
$332 billion is beyond fiscal 2029.
And approximately:
$106 billion doesn’t arrive until after 2031.
Again:
Backlog is not cash.
And backlog seven years from now is not equivalent to revenue next quarter.
The longer the period, the more things can change.
Technology.
Competition.
Customers.
Financing.
Pricing.
Power costs.
Chip efficiency.
AI architectures.
And the economics of computing itself.
Here’s the Question Every Investor Should Ask
Suppose Oracle ultimately receives every dollar of that revenue.
Suppose Google’s AI business grows dramatically.
Suppose Microsoft Copilot becomes enormously successful.
Suppose AWS AI usage explodes.
Suppose Meta builds some of the world’s best AI models.
Does that automatically mean shareholders earn exceptional returns?
No.
The missing variable is:
What did they have to spend to get there?
If a company invests $100 billion and eventually generates $150 billion of attractive incremental profit, fantastic.
If it invests $100 billion and generates $20 billion of incremental economic profit, that is very different.
And if everyone invests simultaneously and competition ultimately drives the price of computing down, the return could be worse still.
The Data Center May Last 30 Years. The Expensive Stuff Inside It Won’t.
Amazon’s Andy Jassy argues that data centers have useful lives exceeding 30 years and can support multiple generations of servers.
That’s a reasonable argument.
But Burry responds with an important distinction.
According to his analysis, the reusable data-center infrastructure represents only approximately 10–15% of the cost of a fully equipped data center.
Much of the remaining investment must be refreshed as servers and chips change.
Think about an airport.
The terminal might last 50 years.
That doesn’t mean the airplanes last 50 years.
The airport is valuable infrastructure.
But airlines still need to keep buying airplanes.
AI data centers have a similar issue.
The building may have a long life.
The GPUs don’t necessarily have the same economic life.
What Happens If AI Keeps Growing — But Growth Slows?
This may be the most important scenario for GYP Members to understand.
The bearish case doesn’t require:
AI demand collapsing.
It doesn’t require:
ChatGPT disappearing.
It doesn’t require:
Microsoft failing.
It doesn’t require:
Nvidia stopping GPU sales.
Imagine AI computing demand is currently growing 80% annually.
Companies build infrastructure expecting something close to extraordinary growth to continue.
Then demand growth slows to:
25%.
That’s still fantastic growth.
Most industries would love 25% growth.
But if the infrastructure was built for 60%, 70% or 80% growth, suddenly there may be too much capacity.
Then:
GPU availability improves.
Cloud-compute prices fall.
Customers negotiate harder.
Data-center utilization falls.
Margins compress.
Older chips lose value faster.
Depreciation assumptions prove too optimistic.
Companies take write-downs.
Return on invested capital falls.
AI didn’t fail.
Expectations failed.
And markets price expectations.
This Is Why Great News Can Eventually Become Bad News
This is the fascinating paradox in Burry’s argument.
Imagine another blockbuster Nvidia earnings report.
Demand exceeds supply.
Microsoft wants more GPUs.
Meta wants more.
Amazon wants more.
Google wants more.
Oracle wants more.
Everyone announces larger capital-expenditure budgets.
The market celebrates.
But a capital-cycle investor may look at exactly the same news and think:
This increases the amount of future supply.
The better today’s profitability becomes, the more competitors invest.
The more they invest, the greater tomorrow’s capacity.
The greater tomorrow’s capacity, the greater the possibility that future returns decline.
That’s why sometimes the most dangerous moment in a capital cycle can occur when the fundamentals appear strongest.
Remember the Dot-Com Bubble: The Internet Was Right
This is an important historical lesson.
The internet wasn’t a fraud.
The bulls were correct about the technology.
The internet changed almost everything.
E-commerce exploded.
Advertising moved online.
Cloud computing emerged.
Streaming replaced traditional media.
Smartphones connected billions of people.
Some of the world’s largest companies were ultimately created by the internet.
And yet investors lost enormous amounts of money during the dot-com collapse.
Why?
Because being right about a technology is not the same thing as being right about:
valuation,
timing,
capital spending,
or
which companies ultimately capture the profits.
Burry sees parallels in today’s capital cycle. His research argues that current net corporate investment relative to GDP is already near extraordinary historical levels, with further AI investment still coming. (Cassandra Unchained)
That does not mean history must repeat.
It means investors should understand the historical pattern.
What Burry Thinks Could Happen Next
Burry isn’t necessarily predicting that the entire AI complex collapses tomorrow.
Quite the opposite.
The spending boom may continue.
Demand may remain extremely strong.
Capital expenditures may continue setting records.
Stocks may continue rising.
But he believes today’s spending creates the possibility of significant write-downs later if the economics don’t justify the enormous amount of capital being deployed.
In his longer analysis, Burry specifically raises 2028–2029 as a period when potential write-offs could become meaningful. (Cassandra Unchained)
That is very different from saying:
“Sell everything Monday morning.”
It means the financial consequences of today’s decisions may not become visible for several years.
The Grow Your Pile Lesson: Price Matters
This brings us back to one of the most important principles in investing.
There is a price at which almost any great company can become a bad investment.
And there is a price at which many mediocre companies can become attractive investments.
A stock isn’t simply:
good company / bad company.
The equation is:
Business + Expectations + Price + Risk = Investment
Microsoft can be an extraordinary company.
Google can dominate search and AI.
Amazon can dominate cloud infrastructure.
Meta can generate enormous cash flows.
Oracle can have record backlog.
Nvidia can sell record numbers of GPUs.
And an investor can still overpay.
That’s the distinction Burry is trying to force investors to consider.
What Should Traders Watch?
Rather than trying to predict the exact top of AI, we believe traders should watch whether the economics begin changing.
Some of the important questions include:
Are capital expenditures continuing to accelerate faster than cash flow?
Is debt issuance increasing?
Are off-balance-sheet commitments continuing to explode?
Is backlog increasingly moving further into the future?
Are cloud-compute prices beginning to decline?
Are GPU shortages becoming GPU availability?
Are customers requiring more financing?
Are depreciation schedules being extended?
Are margins declining despite revenue growth?
Are companies taking impairments or write-downs on older equipment?
Are lenders beginning to demand higher yields to finance AI projects?
That last item is already worth watching. Recent credit-market reporting indicates investors are demanding wider spreads on some AI-related corporate debt as expected issuance grows dramatically. (Reuters)
None of these individually proves that the cycle has turned.
Together, however, they could tell us that the economics underneath the AI boom are changing.
Don’t Confuse a Great Story With a Great Trade
This is particularly relevant for options traders.
We don’t need to decide today whether:
“AI is a bubble.”
That’s probably the wrong question.
Markets can remain expensive much longer than bears expect.
Great companies can become even more expensive.
Momentum can continue.
AI demand can surprise higher.
Instead, we can ask better questions:
What is the market already pricing in?
How much risk are we taking?
How concentrated is our portfolio?
How much buying power are we using?
What happens if volatility suddenly expands?
What happens if the market falls 10%, 20% or 30%?
Are we being adequately compensated for the risk we’re selling?
This is exactly why position sizing matters.
You don’t need to predict the crash to prepare for one.
The Bull Case and Burry’s Warning Can Both Be True
Perhaps the most useful conclusion is that investors don’t have to choose between:
“AI will change the world”
and
“AI investment has become excessive.”
Both can be true.
Electricity changed the world.
Railroads changed the world.
Automobiles changed the world.
Telecommunications changed the world.
The internet changed the world.
But every one of those revolutionary technologies experienced periods in which investors poured too much money into building capacity.
Technology adoption and investment returns are two separate things.
That’s the lesson.
The GYP Bottom Line
Michael Burry’s warning isn’t really:
AI is going to fail.
A better interpretation is:
AI may succeed beyond our imagination, but investors may be paying prices and companies may be committing capital that assume today’s extraordinary economics continue for many years.
That’s a very different warning.
The Big Five hyperscalers are committing extraordinary sums to AI infrastructure.
Some commitments stretch many years into the future.
Some sit outside conventional balance-sheet debt.
Some depend upon customers that themselves require enormous amounts of financing.
Some assets may become technologically obsolete much faster than their physical or accounting lives suggest.
And virtually everybody is expanding simultaneously.
That’s exactly what makes this so interesting.
The danger may not come because AI disappoints.
The danger may come because AI succeeds so spectacularly that everybody spends too much money chasing the opportunity.
Eventually supply catches demand.
Competition increases.
Prices fall.
Margins compress.
Returns on capital decline.
And suddenly investors discover something that financial markets have taught us many times before:
Great technology does not guarantee a great investment.
Great growth does not guarantee a great return.
Great earnings do not justify any price.
And perhaps most importantly:
The price you pay—and the risk you take to participate—still matter.
That doesn’t mean abandoning AI.
It means respecting the capital cycle, respecting valuation, keeping buying power available, managing position size and remembering that markets usually turn before the fundamental story becomes obviously bad.
As Burry himself concludes:
Analysis matters. Patience matters.
For traders and investors, we would add one more:
Risk management matters.
Because you don’t need to know exactly when a cycle ends.
You need to make sure you’re still standing when it does.
Grow Your Pile
This content is provided for educational and informational purposes only and should not be considered investment advice or a recommendation to buy or sell any security, option, or other financial product. The discussion of Michael Burry’s views is a summary and interpretation of his published research and does not imply endorsement of his conclusions. Markets are uncertain, and the scenarios discussed may not occur. Options involve risk and are not suitable for all investors. Past performance is not indicative of future results. Always trade your own account, at your own size, and according to your own risk tolerance.
One thing I deliberately emphasized is the distinction between “AI bubble” and “AI capital-spending bubble.” I think that makes the article considerably stronger. Burry doesn’t need AI to fail for his thesis to work; in fact, spectacular AI results today can encourage the overinvestment that creates the problem later. That is the concept I would make the headline lesson for GYP Members.