Oracle Q1 FY27 — The Customer Became the Lender.

A record, and a shrug

Every headline in this release beat. Revenue of $19,345M grew 29.61%. Cloud infrastructure grew 120.73% to $7,388M — an acceleration off the 93% Oracle printed in Q4, which was itself an acceleration off 84% and 68%. GAAP operating income rose 57.31% to $6,728M. Non-GAAP EPS of $1.92 beat the $1.47 of a year ago by 30.61%. Operating cash flow set a record at $23,103M, up 183.82%. The backlog reached $664 billion. Fiscal 2027 guidance went to at least $90 billion, which is 33.62% growth on a $67,357M base.

The stock closed the reaction day at $150.28, down 1.74% from $152.94. Not a crash — a shrug. That is the more interesting response, because it had already crashed: $150.28 is 54.20% below the fifty-two-week high of $328.15 and 51.19% below the $307.86 close a year earlier. The market spent the last twelve months repricing this name around exactly one question, and the quarter answered it in a way nobody had modelled.

The question, since the June print, has been who ultimately pays for the capex. Oracle's Q4 answer was: Oracle does, with roughly $40 billion of new capital including a $20 billion equity sale. That plan is now complete — 141 million shares issued for net proceeds of $19,909M, about $141.20 a share, 56.97% below the fifty-two-week high. Oracle sold nearly a twentieth of itself into its own drawdown, and the proceeds are 82.47% of the entire $24,140M increase in stockholders' equity this quarter. Earnings were not what built the equity base. Issuance was.

And then, in the same quarter, Oracle found a different lender.

The record cash flow is a loan, and the note says so

Operating cash flow of $23,103M is a genuine record and it is genuinely not operations. Net income supplied $4,760M of it, or 20.60%. Deferred revenues supplied $15,360M, or 66.48%. And $11,363M of that deferred revenue sits on a line that has never appeared in an Oracle cash-flow statement before: "increase in deferred revenues from customer prepayments with significant financing component."

The 10-Q is unusually plain about what this is. "During the first quarter of fiscal 2027, we received $11.4 billion of prepayments from customers that included a significant financing component. No prepayments from customers that included a significant financing component were received during the first quarter of fiscal 2026." A year-ago comparative of zero is the cleanest kind of disclosure — there is no trend to argue about, no mix to decompose. Something began this quarter.

The next sentence is the one that matters. "We determine the discount rate based on a rate that reflects the credit characteristics of the party receiving financing, which is generally consistent with our incremental borrowing rate." Read the subject carefully. In a customer prepayment the customer hands over cash before delivery, so the party receiving financing is Oracle. Oracle is telling you that it priced these contracts off its own cost of debt — which is the definition of a borrowing, written by the borrower, in the borrower's own filing. Accounting agrees: the effects "are reflected in deferred revenues," and deferred revenue is a liability. This is prepaid growth is a customer loan in a cash-flow costume in its purest form to date, and for once the analyst does not have to infer it. The issuer named the instrument.

The costume matters because of where the line lands. A bond issue appears in financing activities and every leverage screen on earth picks it up. This appears in operating activities, where it inflates the single metric — cash from operations — that investors use to decide whether a capital programme is self-funding. Oracle's own free-cash-flow table reports operating cash flow at 485% of net income this quarter, against 187% for fiscal 2026 as a whole. Strip the financing component and operating cash flow is $11,740M, free cash flow is negative $16,759M instead of negative $5,396M, and the quarter looks like what it physically was: $28,499M of concrete and silicon against $19,345M of revenue.

Management is not hiding this. Hilary Maxson put the arithmetic on the call: "Our CapEx for the quarter was $28 billion leading to negative free cash flow of $5 billion," then immediately, "Our net cash CapEx, so net of pre was $18 billion for the quarter" (unverified pending the official transcript). The company has coined its own metric — capital expenditure net of prepayments — and it is the metric the guide is now set in.

The series kills the obvious finding

The easy story in this print is a backlog that stalled. RPO went from $638 billion to $664 billion — an increment of $26.0 billion, against the $85.4 billion Oracle added in Q4. That is a 69.55% collapse in the quarterly flow, and it would make a clean, alarming headline.

It is also wrong, and the reason it is wrong is the discipline this pipeline bought at CXMT's expense: a one-period change is a phase, not a decision, until you have the series. Oracle's RPO increments, quarter by quarter across the last five prints, run $68.0 billion, $29.3 billion, $85.4 billion, $26.0 billion. The second-smallest quarter in that series was immediately followed by the largest. A backlog built from a handful of multi-billion-dollar AI contracts does not accrete smoothly; it steps when a contract signs. Calling $26.0 billion a deceleration requires believing that $29.3 billion was one too, two quarters before the biggest bookings quarter in the company's history.

The series does something better than kill the finding — it points at a different line that actually moved. The share of RPO Oracle expects to convert within twelve months went 10%, 10%, 12%, 12%, 13%. Applied to the balances, that near-term slice went from $76.56B to $86.32B, up 12.75% — nearly three times the 4.08% growth in the backlog itself. The backlog is not stalling. It is aging into the revenue window, which is precisely what you want to see before a conversion year, and it now covers 95.91% of the $90 billion revenue guide.

The lagging line and the leading line

Cash capital expenditure is a receipt. It records a decision taken quarters earlier, when the site was chosen and the order placed. The $28,499M Oracle spent this quarter — 51.20% of everything it spent in all of fiscal 2026, up 235.20% year on year — tells you about fiscal 2025's ambitions, not fiscal 2027's.

The commitment notes are where the live decision sits, and they moved this quarter after three quarters of nothing. The 10-Q discloses "$288 billion of additional lease commitments, substantially all related to data center arrangements, that are generally expected to commence between the second quarter of fiscal 2027 and fiscal 2029 and for terms of fifteen to nineteen years that were not reflected on our condensed consolidated balance sheets." That number ran $99.8 billion, $248 billion, $261 billion, $260 billion across the prior four quarters. It sat flat for three of them — $261 billion to $260 billion is a rounding error — and then added $28 billion in one quarter.

The second commitment line did more. Unconditional purchase obligations went from $13,309M at the fiscal year end to $34,150M, up 156.59% in three months. The composition changed with the size: at the year end these were "primarily related to data center power arrangements"; now they are "primarily related to long-term supply arrangements for purchasing components for cloud infrastructure assets and power supply arrangements for data centers." Oracle's 10-K flagged the pivot as a subsequent event in June — "we entered into an additional $19 billion of unconditional purchase commitments for cloud infrastructure assets that commence in fiscal 2027 and have a term of five years." The company stopped committing only to electricity and started committing to silicon, on a five-year term, before this quarter opened.

So the two readings disagree in the useful direction. The lagging line says the current build is enormous. The leading lines say the next one is contracted and larger.

Fifteen to nineteen years, against five

Put the two commitment books side by side and the structural problem is a calendar, not a number.

On the revenue side, Oracle tells you exactly how long the backlog lasts: 13% within twelve months, 37% across months thirteen to thirty-six, 34% across months thirty-seven to sixty. That is 84% of $664 billion consumed inside five years, leaving 16% — about $106.2B — beyond it. On the cost side, the leases Oracle is signing to serve that backlog run "for terms of fifteen to nineteen years," commencing between Q2 FY27 and fiscal 2029. Those obligations run to roughly 2041 at the near end and 2048 at the far one.

The full obligation stack now totals $491,293M — borrowings of $125,337M, on-balance-sheet operating and finance lease liabilities of $34,621M and $9,185M, the $288,000M of leases not yet commenced, and $34,150M of purchase obligations. That is 73.99% of the backlog it exists to serve, and the not-yet-commenced leases alone are 94.97% of Oracle's $303,259M of total assets. Oracle has contracted to pay rent on a datacentre estate roughly equal in size to its entire balance sheet, and disclosed it in a sentence that begins by saying it is not on the balance sheet.

None of that is fraudulent or even unusual — not-yet-commenced leases belong exactly where Oracle put them. It is a duration statement. Fourteen years after the current backlog is 84% consumed, the rent is still due. Everything therefore rests on renewal at a price that covers a cost base fixed in 2026, by a customer set concentrated enough that the Americas supplied 91.63% of this quarter's entire revenue increase and cloud infrastructure supplied 91.45% of it.

Meanwhile the carrying cost is senior to the shareholder is already visible in the P&L. Depreciation and interest together were $4,584M, up 101.58% against revenue up 29.61%, and they now consume 68.13% of GAAP operating income against 53.17% a year ago. Depreciation alone ran at 12.63% of opening property, plant and equipment; held at that rate against the closing $127,845M it becomes $16,146M a year, against the $24,300M of non-GAAP net income implied by the $8.10 guide on 3,000M shares. The asset base grew 27.90% in three months. The depreciation schedule has not caught up with it yet, and that is a timing difference, not a saving.

Oracle's own guide says this was the peak

The forward number nobody put in a headline is the one the CFO gave on the call: "We continue to anticipate $90 billion to $95 billion in CapEx for the full year with not more than $70 billion in net cash CapEx" (unverified pending the official transcript). Two things fall out of it, and both are checkable.

First, at guide, Oracle will spend more building the asset than it collects for using it. Capital expenditure of $90 billion to $95 billion against revenue of at least $90 billion is 100.00% to 105.56% of revenue. For scale, that ratio was 82.64% across fiscal 2026 and 56.96% in the year-ago quarter. A company of this size has not run a capital programme at parity with its own top line in the modern record of this sector.

Second, the guide implies the quarter just reported was the high-water mark. Q1's $28,499M is 30.00% to 31.67% of the full-year range, leaving $61,501M to $66,501M for three quarters — an average of $20,500M to $22,167M, a sequential step down of 22.22% to 28.07% that then holds flat for the rest of the year. Oracle has quietly guided its own capital spending to fall by roughly a quarter next quarter and stay there, in the same release that reports 850MW of capacity delivered, more than 300,000 GPUs shipped, "almost triple the capacity delivered in Q4 FY26," $30 billion of new contracts booked and a lease book that just grew $28 billion. Those two statements are in tension. One of them will give.

The net-cash ceiling is tighter still. $18,000M in Q1 annualises to $72,000M against a "not more than $70 billion" cap — a 2.86% overshoot in the first quarter, consuming 25.71% of the year's allowance where a straight quarter would take 25.00%. And the external funding share has to fall to make the year work, not rise: Q1 covered $10,499M of gross capex from outside its own cash generation, 36.84%, while the full-year plan implies $20,000M to $25,000M, or 22.22% to 26.32%. The guide only closes if gross spending falls hard. It does not close on more prepayments.

Grading the June call

In June this desk published "The Backlog Compounds, the Funding Bill Arrives" on a quarter that showed RPO at $638 billion, fiscal 2026 capex of $55,663M, free cash flow of negative $23,686M, and a fiscal 2027 plan to raise roughly $40 billion including $20 billion of equity. The call was that funding, not demand, was the binding constraint.

That direction was right and the mechanism was wrong, and the mechanism is the more interesting half. The bill did arrive — the ATM was fully drawn at about $141.20 a share, 56.97% below the high, which is as expensive as equity gets. But the June piece assumed the next increment would come from the same places: more equity, more debt, more preferred. It did not. Borrowings fell $4,204M. The next increment came from the customer, through a line item that did not exist when that piece was written. The correct forward question after Q4 was not "how much more capital will Oracle raise" but "from whom" — and the answer turned out to be the counterparty on the other side of the backlog.

Clay Magouyrk said it directly: "We closed more than $30 billion of additional AI contracts in Q1 without requiring additional capital from Oracle" (unverified pending the official transcript). Maxson gave the structure: "The vast majority of those new contracts were via prepay or bring your own hardware." The press release says the same thing in filing language — "Based on the structuring of those new contracts, the Company confirms there is no incremental impact on its plans to raise capital." Three statements of one fact: the marginal AI contract at Oracle no longer consumes Oracle's balance sheet. That is a genuine change in the business model, and it happened inside one quarter.

The strongest case against this read

The best argument against everything above is that I have described an improvement and called it a risk.

Run it properly. In June, Oracle faced a $664 billion-scale build funded by its own debt and its own equity, and the equity was being sold at half price. A customer prepayment is strictly better than every alternative on the table. It is cheaper than Oracle's own senior notes, because the customer is trading cash for delivery priority rather than for a coupon. It is non-dilutive, unlike the 141 million shares that just went out at $141.20. It carries no covenant, no maturity wall and no refinancing risk, which is the entire argument against every levered AI operator in this coverage. And it is a better demand signal than a backlog number can ever be: RPO is a promise, while $11,363M of wired cash is a customer putting its own balance sheet behind the promise. On that reading, Oracle did not become a borrower — it discovered that its backlog was good enough collateral to make the customer pre-fund the capex, and the right response is to re-rate the backlog's quality, not to discount the cash flow.

The conversion data supports that view rather than mine. The twelve-month recognisable slice grew 12.75% against 4.08% for the backlog as a whole. Gross margin compressed exactly as management said it would, for the reason they gave — Maxson called it "impacts from ramping up our data centers" and infrastructure "a lower gross margin business... much lower R&D and sales associated with it," which the expense lines confirm, with sales and marketing −12.22% and research and development −3.61% against revenue up 29.61%. That is an operator converting a book, not one straining.

Here is where I think it fails, and it is narrow. A prepayment does not remove risk from the system; it relocates it. Oracle's exposure stops being the refinancing market, which is diversified, liquid and continuously priced, and becomes the solvency of the specific counterparties who prepaid — which the filing does not name, does not size individually, and does not disclose concentration for. Oracle swapped a risk everyone can observe for one nobody outside the company can. That is a genuine improvement in cost and a genuine degradation in transparency, and the second only matters if the first is ever tested. But the whole structure assumes the prepayers stay solvent for the fifteen to nineteen years of the leases signed against them, and the backlog they prepaid against is 84% gone in five.

Valuation, and what settles it

At $150.28 on 3,000M diluted shares, Oracle is about $450.84B of market capitalisation and 18.55x the $8.10 guide — a market multiple for a business growing revenue 33.62%. That is the bull case in one number and it is not a silly one. Add the debt and net the cash and enterprise value is roughly $539.10B, or 5.99x the $90 billion guide. Add both lease books, which is the honest treatment of a datacentre operator that leases its estate, and it is roughly $870.91B, or 9.68x. The gap between 5.99x and 9.68x is the whole argument, and it is $331,806M of commitments that a screen will not show you.

So I am not short the demand and I am not paying for the multiple. The position I would hold is the one this desk has held across the complex all year — picks-and-shovels capture over the levered operator — because the marginal dollar of this buildout now reaches a supplier either way, and Oracle's own $34,150M purchase-obligation book, up 156.59% in a quarter, says who collects it.

The dated test is the Q2 FY27 print, guided for December 9, 2026. Three lines decide it, in this order:

• The financing-component line. Does a second tranche arrive? Roughly $10B or more confirms a repeatable funding channel and the $70 billion net-cash ceiling holds. A small number or a blank means Q1 was one contract, and the funding question from June is simply back.
• The imputed interest. Oracle says that in Q1 "such amounts were immaterial" — true, on money that arrived part-way through a quarter. Q2 carries a full quarter of accrual on $11,363M at Oracle's incremental borrowing rate. If it is still immaterial, my reading of the note is wrong and this is nearly free money. If it shows up as a discrete interest line, the cost of the structure becomes visible and every subsequent prepayment gets priced against it.
• Gross capital expenditure. The guide needs $20,500M to $22,167M a quarter from here. A Q2 print at or above Q1's $28,499M breaks the $90 billion to $95 billion range in the first half, and the $70 billion net ceiling with it.

I am wrong if Q2 brings a second large prepayment with still-immaterial interest and capital expenditure steps down into the $20,500M to $22,167M the guide implies. That combination would mean Oracle has found genuinely cheap, repeatable, customer-supplied funding for a build it can pace at will — and the 18.55x is too low. I am right if the prepayment does not repeat, or if it repeats and brings a visible interest cost with it, because then the $23,103M of record operating cash flow was a one-quarter financing event dressed as operations, and the fifteen-to-nineteen-year rent is still due either way.