Oracle's $638 billion remaining performance obligation is being quoted as though it were a valuation. It is an accounting disclosure: contracted revenue not yet recognised, with a schedule attached. The four quarterly readings of fiscal 2026 — $455bn, $523bn, $553bn, $638bn — explain why two different backlog figures circulate in coverage, and set up the comparison that matters: RPO grew 363% year over year while Q4 revenue grew 21%. At Oracle's Q4 run rate the balance represents about eight years of revenue, and roughly half of it is reported to trace to a single counterparty. Moody's has forecast leverage toward 4x EBITDA and S&P has Oracle at BBB-, which is the bear case the backlog headline has to answer.
Oracle reports its first quarter of fiscal 2027 after the close today, 10 September 2026. The number every preview leads with is $638 billion: the company's remaining performance obligations at the end of fiscal 2026, up 363 percent year over year. It is being quoted the way a market capitalisation gets quoted, as though it were a measure of what Oracle is worth. It is not. It is an accounting disclosure with a defined conversion schedule, and understanding the schedule is what separates reading the number from repeating it.
Remaining performance obligation is the revenue a company has contracted for and not yet recognised. It exists because revenue recognition rules require you to book revenue as you deliver the service, not when the customer signs. A ten-year cloud contract signed today produces a large RPO balance today and a thin trickle of revenue for a decade. The balance is a promise with a clock attached.
Three things follow from that definition, and all three are lost when the figure is quoted alone. RPO is not cash: no money has changed hands for the undelivered portion. RPO is not funded: the customer's obligation to pay depends on the customer continuing to exist and continuing to want the capacity. And RPO is an aggregate, which means a single line conceals whether the balance came from ten thousand customers or from two.
Coverage of Oracle currently circulates at least two backlog figures — $523 billion and $638 billion — usually without saying which period each belongs to, which makes the discrepancy look like a dispute. It is not. Oracle's own quarterly disclosures through fiscal 2026 read:
The $523 billion figure is the Q2 reading, which was itself announced as a record at the time. The $638 billion is the fiscal year-end position. Anyone citing the smaller number today is nine months out of date; anyone treating the pair as a contradiction has not checked which quarter they came from. The sequence is more informative than either endpoint, because it shows the balance building in large uneven steps — $68 billion, then $30 billion, then $85 billion — which is the signature of a handful of very large contracts landing, not of broad-based demand accumulating.
Set the backlog growth against the revenue growth in the same release. In the fourth quarter of fiscal 2026 Oracle reported total revenue of $19.2 billion, up 21 percent, with total cloud revenue of $9.9 billion, up 47 percent, and cloud infrastructure revenue of $5.8 billion, up 93 percent. Those are strong numbers. They are also an order of magnitude away from 363 percent.
That gap is not a discrepancy to be explained away; it is the mechanism working as designed. A contract signed in fiscal 2026 for capacity to be delivered over the following decade adds its full value to RPO immediately and almost nothing to revenue immediately. RPO growth of 363 percent therefore cannot translate into revenue growth of a similar order in any near quarter, and any commentary that implies it can has misread what the balance is.
Divide it out. At the Q4 revenue run rate — $19.2 billion a quarter, roughly $77 billion annualised — $638 billion of contracted obligation represents about eight years of revenue at current volumes. That is the honest way to hold the number: not a valuation, but a statement that Oracle has booked roughly eight years of its present-size business, to be delivered by a company that has to build most of the capacity first.
Several previews cite management's expectation that roughly 12 percent of the balance converts to revenue within twelve months. That figure is worth understanding and worth caveating in the same breath. Twelve percent of $638 billion is about $77 billion, which is almost exactly Oracle's current annualised revenue run rate — and Street expectations for the quarter reporting today are around $19.1 billion of revenue and $1.74 in earnings per share, with cloud revenue guided to grow 58 to 64 percent. So the 12 percent figure is at least internally consistent with what analysts expect: it implies that nearly all of Oracle's near-term revenue is backlog conversion rather than new business, which is what you would expect of a contracted, subscription-shaped book.
We could not confirm the 12 percent figure in Oracle's Q4 fiscal 2026 earnings release, which discloses the RPO balance and its growth but not a near-term conversion percentage. It is reported as management guidance rather than as a disclosed number, and readers who want it verified should look for the remaining performance obligations note in the fiscal 2026 Form 10-K, where the recognition schedule is required to appear. Nothing above depends on it: the eight-years-of-revenue arithmetic uses only figures Oracle published itself.
The concentration question is the one the single number is worst at answering. Analyst estimates put roughly half of Oracle's remaining performance obligations against a single counterparty, OpenAI. The underlying agreement is reported to cover on the order of 4.5 gigawatts of capacity, with the Abilene, Texas build reported at around 400,000 Nvidia GB200 chips and roughly $40 billion. These are reported figures about a private contract between two companies, not disclosed terms, and should be read as such.
But the structural point survives whatever the exact split turns out to be. A backlog concentrated in a small number of AI counterparties carries a different risk profile from the same balance spread across thousands of enterprise customers, and the aggregate line reports the two identically. Enterprise software backlogs have historically been dull precisely because they were diversified. This one is not.
There is a second constraint that the word backlog obscures. Oracle cannot bill for capacity it has not built. Gigawatts of contracted compute require data centres, power interconnection and chips, all of which arrive on their own schedule and none of which is inside Oracle's control. The path from RPO to revenue runs through construction, and construction is where the balance is most likely to slip.
Moody's has forecast that Oracle's debt will grow faster than its EBITDA, taking leverage toward 4 times before earnings begin to outpace borrowing, and has held a negative outlook on the credit. S&P has Oracle at BBB-, one notch above sub-investment grade, with adjusted leverage projected into the mid-4x area in fiscal 2027 — past where S&P considers a standard BBB profile comfortable. Barclays has raised the possibility of the debt approaching junk status. Oracle raised roughly $30 billion in debt and equity financing in early 2026 against capital expenditure guidance near $50 billion.
That is the counter-argument the bull case has to answer, and it is not a sentiment argument. It says the company is borrowing at scale to build the capacity that converts the backlog, and that the ratings agencies think the borrowing is running ahead of the earnings. If the backlog converts on schedule, the leverage is transitional. If it slips, the leverage is not.
Which is why the useful thing to read tonight is not the backlog headline. It is the capital expenditure line, the cloud infrastructure growth rate against the 58 to 64 percent guidance, and any change in the RPO recognition schedule. Those three tell you whether the eight years of contracted revenue is arriving. The $638 billion, by itself, only tells you it was promised.
No. Remaining performance obligation is revenue a company has contracted for but not yet earned, because accounting rules require revenue to be recognised as the service is delivered rather than when the contract is signed. It is not cash, it is not funded, and it converts into revenue over the life of the contracts — in Oracle's case, over many years.
We take apart the disclosures behind the headlines — what they measure, what they omit, and what they cannot tell you.
More Business analysisFour underwriters retained, nothing filed. The hard part of a DeepSeek listing is not the valuation — it is what an open-weights company writes in the revenue section of a prospectus.
ISS and Glass Lewis both say reject it. Neither recommendation can change the result. At a company where one holder controls 86.4% of the stock, a say-on-pay vote stops being a decision and becomes a disclosure instrument.
The headline was the 94% drawdown from a $3.44 billion peak. The more useful arithmetic is the cash sitting inside the thing being bought — and the fact that everyone except the angels was paid in another private company's paper.