Oracle Has $664 Billion in Future Contracts. So Why Are Investors Worried?
Oracle has a number that looks almost unbelievable.
$664 billion.
That's how much the company reported in remaining performance obligations, or RPO, at the end of its latest quarter.
In simple terms, it represents revenue Oracle has contracted but has not yet recognized.
For comparison, Oracle generated:
$19.3 billion in revenue during the quarter.
So the obvious question is:
If customers have already committed hundreds of billions of dollars, what is there to worry about?
The answer is simple.
A contract is not profit.
And in the AI infrastructure business, there can be an enormous amount of spending between signing a contract and earning money from it.
Start With a $10 Billion Contract
Imagine Oracle signs a long-term AI cloud contract worth:
$10 billion.
It sounds like Oracle just made $10 billion.
It didn't.
The customer may consume computing capacity over several years.
Oracle recognizes revenue as it provides that service.
But first, Oracle has to be able to provide it.
That may require:
GPUs.
Servers.
Networking equipment.
Cooling systems.
Data centers.
Power infrastructure.
And enormous amounts of electricity.
So the real sequence looks more like this:
Customer signs contract
↓
Oracle needs more computing capacity
↓
Oracle spends billions building it
↓
Customer begins using the capacity
↓
Revenue is recognized over time
The contract can arrive first.
Much of the cash required to fulfill it can leave much earlier than the revenue arrives.
Look at Oracle's Latest Quarter
Oracle's latest numbers make this unusually easy to see.
In its fiscal first quarter of 2027, Oracle reported:
$19.3 billion — total revenue
$11.6 billion — cloud revenue
$7.4 billion — cloud infrastructure revenue
Oracle Cloud Infrastructure, or OCI, grew:
121% year over year.
Demand is clearly growing.
Oracle said it also delivered 850 megawatts of additional data-center capacity during the quarter.
But now look at another number.
Oracle's capital expenditures for the quarter were:
$28.5 billion.
That's almost $9 billion more than the company's entire quarterly revenue.
The year before, quarterly capital expenditure had been about $8.5 billion.
This is the strange economics of the AI infrastructure boom.
More demand can require enormous spending before it produces all of the expected revenue.
Free Cash Flow Shows the Other Side
Oracle remained profitable.
Revenue grew.
Cloud infrastructure more than doubled.
But free cash flow for the quarter was approximately:
-$5.4 billion.
That doesn't mean Oracle lost $5.4 billion in accounting profit.
Free cash flow asks a different question:
After the cash generated by the business and the money spent on capital investment, how much cash was left?
For Oracle, massive infrastructure spending pushed that number negative.
That's why two investors can look at the same company and see two very different things.
One sees:
$664 billion in future contracted revenue.
The other sees:
the enormous amount of infrastructure required to deliver it.
Both are real.
So What Exactly Is the $664 Billion?
Now we can give the number its proper name.
Remaining Performance Obligations.
RPO represents contracted revenue that has not yet been recognized.
Oracle's RPO reached:
$664 billion
at the end of Q1 FY27.
That was an increase of:
$209 billion from a year earlier.
Oracle also said it signed more than $30 billion of additional AI cloud contracts during the quarter.
That's a powerful sign of future demand.
But RPO is not:
cash sitting in Oracle's bank account,
current-quarter revenue,
or profit.
It is a promise Oracle still has to fulfill.
And fulfilling an AI infrastructure promise can be extremely expensive.
Oracle May Spend Up to $95 Billion This Year
Oracle expects fiscal 2027 capital expenditures of approximately:
$90 billion to $95 billion.
Think about what that means.
Oracle isn't simply selling more software licenses.
It is building physical capacity.
The company needs enough infrastructure to serve the computing demand represented by those giant contracts.
That means investors increasingly have to watch numbers that traditionally weren't the first thing people associated with Oracle:
data-center capacity,
GPU deployments,
electricity,
capital expenditure,
debt,
cash flow,
utilization,
and infrastructure margins.
Oracle's problem isn't finding customers.
The backlog suggests there are plenty.
The harder question is:
How much will it cost to serve them?
Sometimes the Customer Helps Pay for the GPUs
This is where the structure becomes more interesting.
Oracle doesn't necessarily have to buy every GPU with its own money.
The company has disclosed that some large AI agreements include arrangements where customers either:
prepay Oracle for GPUs
or
purchase GPUs themselves and provide them to Oracle.
At the end of Oracle's previous fiscal year, the prepaid and customer-supplied hardware portions of these contracts totaled about:
$75 billion.
Why would Oracle structure contracts this way?
Because GPUs are expensive.
If the customer helps provide or finance the hardware, Oracle doesn't have to raise as much capital itself to build the infrastructure required for that customer.
The relationship changes from:
Oracle buys infrastructure → customer rents computing
to something closer to:
Customer wants huge computing capacity
↓
Customer helps fund or supply GPUs
↓
Oracle provides the infrastructure around them
↓
Customer consumes the computing capacity
The customer isn't merely buying the service.
In some cases, it is helping finance the machinery required to create that service.
But Oracle Still Needs Enormous Amounts of Capital
That doesn't make the financing problem disappear.
During fiscal 2026, Oracle disclosed that it raised:
$43 billion in debt
and
$5 billion in equity.
For fiscal 2027, Oracle had also outlined plans to raise roughly another:
$40 billion
through debt and equity.
And during the latest quarter, the company completed a:
$20 billion at-the-market equity offering
as part of its capital investment program.
So the AI expansion isn't simply:
sell more cloud → make more money.
It looks more like:
sign contracts
↓
finance infrastructure
↓
buy or receive GPUs
↓
build capacity
↓
deliver computing
↓
recognize revenue
↓
find out what margin remains
That last step matters.
Two $10 Billion Contracts Can Be Very Different
Imagine two companies.
Company A
Signs a $10 billion contract.
Needs $1 billion of additional investment to fulfill it.
Company B
Signs a $10 billion contract.
Needs $8 billion of additional investment to fulfill it.
The headline for both could say:
Company Signs $10 Billion Contract.
But those aren't economically identical contracts.
The contract value tells you about demand.
It doesn't tell you how much money the company ultimately keeps.
That's why the next question after seeing a giant AI contract should be:
What must the company spend to fulfill it?
Software and AI Infrastructure Are Different Businesses
This matters especially for Oracle.
For decades, Oracle was known primarily for databases and enterprise software.
Software can have extraordinary economics.
A company spends money developing the product.
Then it can sell versions of that software to many customers without building a new physical facility for each one.
AI infrastructure is different.
A giant new customer may require:
more GPUs,
more servers,
more cooling,
more networking,
more electricity,
and more data-center capacity.
In other words:
More customers can require more physical capital.
Oracle is still a huge software company.
But as OCI becomes a larger part of its growth story, investors increasingly have to understand something else too:
infrastructure economics.
Oracle Is Actually Delivering
None of this means Oracle's AI contracts are merely promises on paper.
OCI revenue increased 121% year over year in the latest quarter.
Oracle said it had delivered more than 300,000 GPUs to AI Cloud customers since the end of the previous quarter.
The company also said it delivered nearly three times as much capacity as it had in Q4 FY26.
Total revenue grew 30%.
So the central question isn't:
Is there real AI demand?
There clearly is.
The question is:
What will that demand be worth after Oracle pays to serve it?
Larry Ellison Added One More Headline
There was also an interesting piece of timing.
Larry Ellison had previously adopted a trading plan that could have allowed him to sell as many as 50 million Oracle shares, worth roughly $7.5 billion at recent prices.
On September 12, Oracle said Ellison had terminated the plan.
No shares had been sold under it.
Oracle said Ellison currently had no plans to sell Oracle stock.
That doesn't tell us whether Oracle shares are cheap or expensive.
And it doesn't answer the infrastructure economics question.
But it arrived while investors were already trying to understand an unusual combination:
enormous AI demand
and
enormous capital requirements.
The Biggest Contract Isn't Necessarily the Best Contract
This is the broader lesson.
AI infrastructure companies are announcing enormous numbers.
Billions of dollars of contracts.
Gigawatts of capacity.
Hundreds of thousands of GPUs.
Huge backlogs.
Those numbers matter.
But every giant contract creates a second question:
What does the company have to spend to fulfill it?
That applies to Oracle.
It applies to cloud providers.
It applies to data-center developers.
It applies to energy companies building infrastructure for AI.
And it applies to the financial institutions funding the entire buildout.
A giant backlog can be excellent news.
It can also be an extremely expensive promise to keep.
Oracle's $664 billion tells us customers want an extraordinary amount of computing.
It does not tell us how profitable those contracts will ultimately be.
For that, we need to follow the money from:
contract
to
GPU
to
data center
to
electricity
to
financing
and finally to:
cash flow.
That's the distance between a giant contract and a great business.
BEYOND THE OBVIOUS.
Sources