Oracle's $638 Billion Backlog: the CFO Just Explained Why the Promise Isn't the Money

Oracle reports first-quarter numbers after the close today, and the stock is already pricing in the verdict. ORCL fell about 5% into the print to roughly $153, more than 50% below its all-time high near $330 — a remarkable thing for a company that keeps reporting record sales and a record backlog.

The figure in every Oracle headline is that backlog: remaining performance obligations (RPO) of $638 billion at the end of May, up 363% from a year earlier and equal to nearly ten times annual revenue. A business with that many signed contracts should be a pure growth story with no demand worries. That is not what has been scaring investors. What scares them is everything the backlog promise does not deliver.

RPO is the accounting name for revenue Oracle has been contracted to earn but hasn't yet. It is a signed agreement, not cash and not profit. How quickly it becomes recognized revenue is the entire question, and new CFO Hilary Maxson has been unusually blunt about it. Of that $638 billion, Oracle expects only about 12% to convert into recognized revenue within the next twelve months, and roughly a third within three years. Most of the backlog is years away.

That lag is not a demand problem. It is a construction-and-financing problem. The bulk of these reservations are multi-year AI data-center deals where this quarter's money must be spent years before a customer's servers are even powered up.

Measure the cost of converting the promise and the picture snaps into focus. In fiscal 2026, the year ended May 31, Oracle spent $55.7 billion on data centers — more than 2.6 times the $21.2 billion it spent a year earlier. It generated about $32 billion in operating cash flow, so free cash flow came in at negative $23.7 billion, flipping from about breakeven a year earlier. The seller of a $638 billion backlog burned roughly $1.74 in building for every $1 its operations threw off.

The backlog is a revenue promise on a credit card.

Oracle does not pretend otherwise. Maxson guided net data-center spending to roughly $70 billion for the current fiscal year and said the company expects to raise about $40 billion more through debt and stock sales — on top of the roughly $48 billion it raised last year. S&P Global has already downgraded Oracle's credit one notch to BBB-, just above junk, citing the scale of the buildout.

Then there is the question of who is paying. S&P estimates OpenAI sits behind roughly half of the $638 billion backlog. Half of a ten-times-revenue promise resting on one customer, whose own commitments depend on its ability to keep raising capital, is the kind of concentration that turns a backlog into a risk rather than an asset. Customers prepaying or supplying their own hardware — about $75 billion worth — ease Oracle's near-term bill, but they also show how much of the "bookings" is infrastructure the buyer is financing itself.

Here is the honest version of what the CFO's tempered message means. The size of the backlog was never the problem; the conversion is. Cloud infrastructure revenue grew 93% in the latest reported quarter while free cash flow stays deeply negative, because revenue recognition lags years of up-front spending. The market has re-priced the stock from about $330 to about $153 to reflect that the headline number is a revenue promise financed increasingly on credit — not a profit stream already in hand.

None of this makes the backlog worthless. If Oracle converts the contracts on schedule, the growth on the income statement would eventually justify the price, and the stock would look cheap in hindsight. If delivery stalls or a customer the size of OpenAI cannot fund its own commitments, the leverage cuts the other way at triple-digit-billion scale. The CFO's warning is the tell: the market is not underweighting the demand, it is demanding proof that the backlog becomes cash before it will believe the price. That proof arrives in twelve-month slices — and today's report is the first one investors get to see.