Summary

  • In the quarter Oracle calls Q1 FY27, capital expenditures were $28,499 million, but a non-GAAP bridge deducted $11,363 million of "customer prepayments with significant financing component for capital expenditures" to reach a net cash outlay of $17,966 million. That deduction was nil a year earlier.
  • Management said the majority of new bookings arrived through prepayment or bring-your-own-hardware terms that "will not require incremental capital from Oracle," while Oracle itself still carried $130,105 million of gross notes payable and $260,000 million of not-yet-commenced data-centre lease commitments at the end of FY2026.
  • The backlog converts slowly: Oracle's Q1 FY27 filing schedules roughly 13% of the $664 billion in remaining performance obligations as revenue within twelve months, against leases whose weighted-average remaining terms run about 12 years for operating and 14 years for finance leases.
  • Customer funding is a real reduction in the cash Oracle must front, but it is not a transfer of the residual risk. The disclosed structure leaves open who prepaid, on what repayment or termination terms, and how concentrated the backlog is.

For a year the argument about Oracle's AI business has been about size: how much capacity has been sold, how many gigawatts announced, how many GPUs delivered. The first quarter of Oracle's fiscal 2027 offers a sharper question, because the company's own disclosures now separate two things that used to be discussed together — the demand it has contracted and the cash it must find to serve it.

In the quarter ended 31 August 2026, which Oracle released on 10 September, total revenue was $19.3 billion, up 30%, and cloud infrastructure revenue was $7.4 billion, up 121% in dollars (Oracle Q1 FY27 release). Remaining performance obligations, the contracted backlog yet to be recognised as revenue, reached $664 billion, $209 billion above the same point a year earlier, after the company booked more than $30 billion of additional AI cloud contracts in the quarter. Those figures circulated widely and are mirrored in the wire distribution of the same release (PR Newswire copy).

The number that moved

The more informative line sits in the cash flow presentation rather than the headline. Oracle reports a non-GAAP measure it calls "net cash outlay for capital expenditures," defined as capital expenditures less other short-term financing cash flow related to capital expenditures and less "customer prepayments with significant financing component for capital expenditures." In Q1 FY27, gross capital expenditures were $28,499 million, against $8,502 million a year earlier. The bridge subtracted $830 million of other short-term financing and $11,363 million of those customer prepayments, giving a net cash outlay of $17,966 million (Oracle Q1 FY27 release).

The same $11,363 million appears in the cash flow statement as an increase in deferred revenues from customer prepayments with a significant financing component. In the year-ago quarter that line was zero. Over the quarter, GAAP operating cash flow was $23,103 million — a record — while free cash flow was negative $5,396 million, and Oracle completed the sale of $20 billion of common stock through an at-the-market program, netting $19,909 million (Oracle Q1 FY27 release).

Read plainly: a material and previously absent category of customer cash entered Oracle's operating cash flow, and the company chose to show readers what capex would look like without it. That is a disclosure decision, not a hidden one. But it also means the gap between what Oracle spent and what Oracle had to find can now be closed, in part, by its customers.

Three ways a customer can carry the cost

On the earnings call, CFO Hilary Maxson said remaining performance obligations rose $26 billion from the prior quarter and that "the vast majority of those new contracts were via prepay or bring your own hardware or similar mechanic, so will not require incremental capital from Oracle," adding that the new backlog "will not impact our CapEx or revenues until fiscal 2028 or beyond" (earnings call transcript). Oracle's own slides carry the same framing, stating that a majority of the quarter's RPO growth came via pre-pay or bring-your-own-hardware (management slides).

Later in the same call, co-CEO Clay Magouyrk refined the claim, describing three distinct mechanisms rather than one: supplier financing arrangements that let Oracle pay for capacity as customers pay it; customers who buy the hardware themselves while using Oracle's operational and cloud capabilities; and customers who pay upfront as a prepayment (call transcript). He also narrowed the language, saying the contracts "don't require additional cash from Oracle" — a treasurer's statement, not a capital-structure one. Hardware still gets bought. Capacity still gets built. The question is whose balance sheet carries it while it is being built.

Those three mechanisms are economically quite different, and Oracle does not split them. Supplier financing shifts the timing of Oracle's own payments. Bring-your-own-hardware shifts ownership of the depreciating asset to the customer. A prepayment shifts cash timing to Oracle before delivery. Only the first is a pure timing device; the other two change who owns the asset and who bears its residual value.

What Oracle still carries itself

Whatever customers fund, a large part of the build-out remains on Oracle's own accounts, and the FY2026 annual report, covering the year ended 31 May 2026, quantifies it. Notes payable and other borrowings were $130,105 million gross, or $129,541 million net, up from $92,917 million gross a year earlier, with $7,199 million current and $122,342 million non-current (notes payable detail). Operating lease right-of-use assets were $29,690 million, operating lease liabilities $3,542 million current and $26,648 million non-current, and finance lease liabilities $7,701 million, with weighted-average remaining terms of roughly 12 years for operating and 14 years for finance leases (leases detail).

The largest single item is not on the balance sheet yet. Oracle disclosed $260,000 million of additional operating lease commitments, substantially all of them data-centre arrangements that had not commenced, generally expected to begin between the fourth quarter of FY2026 and fiscal 2028 and to run for fifteen to nineteen years (leases detail). It also disclosed $13,309 million of unconditional purchase commitments at year-end plus a subsequent $19,000 million commitment over five years, and a guarantee of up to $3,300 million of a lessor's borrowing maturing in September 2026. The FY2026 annual report states that the prepaid and customer-supplied-hardware portions of its large AI contracts then totalled $75 billion and that fiscal 2026 funding included $43 billion of debt and $5 billion of equity, with fiscal 2027 expected to raise roughly $40 billion through combined debt and equity including the $20 billion at-the-market program (FY2026 results release).

So the accurate description is not that customers fund Oracle's build-out. It is that customers fund an increasing share of it while Oracle simultaneously commits to a quarter-trillion dollars of long-dated leases and adds roughly $37 billion of gross borrowings in a single year. Those are two different exposures with two different risk owners.

A 13% conversion rate against fifteen-year leases

If customer funding reduces the cash Oracle must find, the remaining question is whether the revenue arrives before the obligations do. The Q1 FY27 quarterly report gives the schedule: of the $664 billion in remaining performance obligations, approximately 13% is expected to be recognised as revenue over the next twelve months, 37% over months 13 to 36, 34% over months 37 to 60, and the remainder thereafter (10-Q RPO schedule, as reported). Management separately said it expects around half of the backlog to convert to sales within 36 months (earnings call transcript). The quarterly report's RPO figure also uses the standard practical expedient that excludes variable consideration allocated to wholly unsatisfied obligations — a detail that matters as more contracts are written on prepaid or customer-supplied-hardware terms (10-Q analysis).

Set that against the lease side, where the terms run fifteen to nineteen years for the not-yet-commenced arrangements and about twelve years remaining on the operating leases already recognised. Oracle's own risk disclosure names the mismatch directly: the terms, renewal options and pricing adjustments in its long-term data-centre leases "typically do not align with the duration and pricing of customer contracts," and if customers do not renew, Oracle may be unable to re-lease, repurpose or assign the capacity on acceptable terms (FY2026 10-K). The same filing warns that if Oracle overestimates demand or a key customer cannot pay, it could be locked into multi-year commitments for excess data-centre space and related capital expenditures, as well as associated financings, without corresponding revenue — and it notes that prepayments under some committed contracts are a credit-risk tool that cannot fully eliminate customer credit risk.

That is the crux of the funding question. A customer prepayment removes an immediate cash need. It does not shorten a fifteen-year lease, and it does not by itself prove that the capacity being financed will be re-let at a profit if the customer that funded it walks away at renewal.

Concentration: disclosed at the revenue line, silent at the backlog

Oracle's FY2026 annual report states that no single customer accounted for 10% or more of total revenues in fiscal 2026, 2025 or 2024 — a comforting revenue-level figure (10-K analysis). The same filing warns separately that in certain OCI offerings Oracle is "more concentrated among a number of large customers," which is a statement about the contracted future, not about past revenue. Neither the earnings release nor the quarterly report identifies which counterparties prepaid the $11.363 billion or supplied hardware, and Oracle does not break out remaining performance obligations by customer (10-Q analysis).

Third-party reporting has noted a previously reported OpenAI agreement to purchase approximately $300 billion of Oracle computing capacity over about five years, which Oracle has not broken out in its Q1 disclosure (sector reporting). If a small number of large counterparties supply most of the prepayments and most of the backlog, then the customer-funded model is not a broad-based market pricing Oracle's capacity — it is a concentrated credit exposure dressed as demand, and the difference is only visible when those customers renew or fail to.

What else the quarter showed

To be clear about what is not in dispute: Oracle delivered 850 megawatts of additional data-centre capacity and more than 300,000 GPUs to AI cloud customers since the end of the prior quarter, with AI infrastructure GPU utilisation of 97.9% and renewed or resold four-year-old capacity earning an average premium of about 20% (datacentre reporting). Those are real operating facts, and they are the strongest available counterweight to a purely financial reading of the quarter: the installed fleet is being re-let rather than abandoned.

They do not, however, tell us who owns that fleet. A GPU that a customer bought and Oracle operates, and a GPU Oracle bought and leases to a customer, produce the same utilisation statistic and radically different balance-sheet consequences. Until Oracle splits its disclosure along that line, utilisation and the premium are evidence about operations, not about who carries the capital risk.

What would falsify the customer-funded story

Four observable developments would materially change this reading.

First, gross capex rising while the customer-prepayment deduction falls back toward nil. That would show the customer-funded share is not scaling with the backlog — that prepayments were a one-quarter feature, not a structural shift.

Second, disclosure of prepayment or customer-supplied-hardware counterparties, or of backlog concentration, that shows the customer-funded portion resting on one or two names. That would convert an apparent diversification into a single-credit dependency.

Third, commencement of the $260 billion of not-yet-commenced lease commitments onto the balance sheet without matching revenue recognition. That would make the duration mismatch visible in reported results rather than prospective risk factors.

Fourth, remaining performance obligations converting materially below the disclosed 13%-within-twelve-months and 50%-within-36-months schedule across successive quarters. And beneath all four: any disclosure that customer prepayments carry repayment, termination or refund terms that make them economically closer to financing than to demand.

The reverse is equally observable. If the customer-prepayment deduction keeps pace with gross capex across the next several quarters, if the backlog converts on schedule, and if Oracle names or at least de-concentrates the counterparties, then customer-funded capacity is a durable substitute for Oracle-funded capacity — and the market's worry is misplaced. Until one of those becomes visible, the honest position is that Oracle has partially reallocated the cash cost of the build-out without disclosing who now holds the residual risk.

For readers tracking Oracle through its directory entry, the practical instruction is to read three lines together each quarter: the net cash outlay bridge, the RPO conversion schedule, and the lease-commitment disclosure. Separately they describe a company growing fast, a backlog that is long-dated, and a landlord with a long memory. Together they describe who is paying. The directory entry remains the reference point for the company itself.