Oracle Proved the AI Boom Is Real. Can It Afford to Build It?
A 638 billion dollar backlog settled the demand question for good. A 23.7 billion dollar cash burn opened a harder one. Here is what the filing shows, and the test we set before it landed.
For two years, the entire argument about artificial intelligence came down to one question. Was the demand real, or was the market paying for a story.
Last week, Oracle answered it. The answer was so loud it should have ended the debate on the spot.
It did not. Because sitting in the same report was a second number that asks a harder question, and almost nobody is talking about it.
The number that ended the argument
Oracle reported 638 billion dollars in remaining performance obligations. That is the plain term for orders already signed and not yet delivered. It grew 363 percent in a single year. In the most recent quarter alone, it rose by 85 billion dollars.
Numbers that large invite suspicion, so look at what sits underneath them. A large block of that backlog is prepaid. Customers are handing Oracle cash in advance. In some cases they are buying the chips themselves and shipping them in for Oracle to run.
That detail matters more than the headline figure, and here is the mechanism. An ordinary order can be cancelled, delayed, or renegotiated. A prepayment cannot. When a customer pays before the product exists, they are not expressing optimism. They are expressing fear of being last in line for scarce computing power.
That is the strongest demand signal a company can produce. Not a forecast. A deposit.
The one claim everything rested on
This is the part we care about most, because anyone can narrate a quarter after it happens.
Before Oracle reported, we ran it through The Stock Story Firewall. It does one thing. It strips a company down to the single claim its valuation depends on, the one that takes the whole story down with it if it breaks.
For Oracle, that claim was this. The AI demand behind the stock is real and durable, it is locked into contracts Oracle can actually collect, and Oracle is physically building the capacity to deliver it. Not a one-quarter spike. Not paper bookings dressed up as growth. Real, contracted, and under construction.
That claim only holds if four things are true at the same time. So before the print, we wrote down four lines and said exactly what would prove each.
Demand had to still be accelerating, not fading. We required cloud infrastructure growth above 50 percent.
The demand had to be contracted, not hoped for. We required a backlog above 100 billion dollars, growing at least 40 percent.
The legacy database business, the engine that pays for all of this, could not be eroding underneath. We required its support revenue to hold flat or grow.
And Oracle had to be spending real money to build, not just booking promises. We required at least 4 billion dollars of capital spending in the quarter.
Four conditions, one claim. Here is what the filing returned.
Cloud infrastructure grew 93 percent, accelerating through the year from 55 to 68 to 84 to 93. The backlog reached 638 billion dollars, up 363 percent. The legacy support line held roughly flat, up about 1 percent, while cloud grew 47 percent on top of it. And capital spending ran near 56 billion dollars across the year, far past the bar.
Four conditions. Four clears. The claim did not squeak through. It held on every count.
Then we read the cash flow statement
That last number is where the story turns. The fourth condition, real money spent on real construction, is exactly what proved the claim. It is also what created the new problem.
To deliver this backlog, Oracle spent about 56 billion dollars building last year, up from 21 billion the year before. Its free cash flow, the cash left after running the business and funding the buildout, came in at negative 23.7 billion dollars. To cover the gap, it raised tens of billions in new debt and equity.
Read those two facts together. Oracle has the most certain demand in its history. It is also spending far more than it earns to meet that demand, and borrowing the difference.
If you read the headline and stopped, you would file Oracle as a sure thing. The filing says something more interesting, and more useful. The question that consumed two years, will anyone buy this, is settled. The question that matters now is different. Can Oracle fund and physically build everything it has already sold, without the economics breaking on the way.
Demand was the easy part.
The bull case is better than it looks
It would be lazy to stop on the scary number, so here is the strongest version of the other side.
This cash burn is not reckless spending into a hope. It is spending against signed, often prepaid, contracts. The customers are pre-funding the buildout. The backlog gives Oracle three to four years of revenue visibility, which is exactly the cover a company wants before it leans this hard on debt.
Seen that way, negative free cash flow is not a wound. It is a decision to pull forward as much certain demand as the balance sheet can carry. If the contracts are real, and the prepayments say they are, the spending is the rational move.
That case is sound. It is also incomplete.
Demand was never going to be the risk
Here is what the bull case quietly assumes, and what the filing does not yet show.
A signed contract tells you revenue is coming. It tells you nothing about the margin on that revenue. A backlog of 638 billion dollars at thin margins is a very different business than the same backlog at fat ones, and Oracle is not breaking that out.
A prepayment tells you the customer is committed. It does not tell you Oracle can secure the power, the chips, and the construction to deliver on schedule. Execution at this scale has broken stronger operators than this.
And debt raised in a high-rate world is cheap insurance right up until a single large customer slips.
None of these are predictions. They are the specific things that are now unknown. And they are unknown because the demand question is no longer where the risk lives. The risk moved. It moved off the order book and onto the balance sheet and the build.
What we will be watching now
Our call cleared on every line, so the verdict is simple. The AI demand Oracle sells is real, proven, and still accelerating. That part is no longer in doubt.
The evidence that mattered most was not the size of the backlog. It was the prepayments, which made the demand undeniable, and the negative free cash flow, which told us where the next fight gets fought.
What we cannot yet see is the margin on that backlog and Oracle’s ability to deliver it on time. Those are the numbers we will be reading next quarter, while the rest of the market re-reads the backlog.
And here is the idea worth keeping, the one that outlasts this single company. When a business proves explosive demand, the next question is always the cost of meeting it. Demand is the easy half. The cash flow statement is where hypergrowth either compounds or cracks. Celebrate the order book if you want. Then turn the page and read what it cost.
Tomorrow we go before the crowd, to a company almost no one is watching and the headlines never mention. If you want the reveal the moment it lands, along with the rest of this week’s verdicts, subscribe and read with us.
Not investment advice. The subscriber decides.