On Thursday evening Oracle published results for the first quarter of its fiscal 2027, and at first reading it was the quarter the market had been waiting a year for. Revenue rose 30 percent to 19.35 billion dollars, ahead of the 19.14 billion analysts had pencilled in. Adjusted earnings per share came in at 1.92 dollars against a consensus of 1.74. Cloud infrastructure revenue, the money Oracle makes from the data centers it builds for AI customers, grew 121 percent to 7.4 billion dollars, accelerating from 93 percent in the prior quarter. Remaining performance obligations, the contracted backlog, reached 664 billion dollars. And full-year guidance went up: at least 90 billion dollars of revenue, growth of 34 percent in constant currency, where June had promised 27 to 29 percent.
The stock had lost 5.4 percent in the regular session, closing at 152.94 dollars, on a day when the bond market sold off anything with a large appetite for capital, and it recovered four to six percent of that after hours depending on the hour. Net of both moves, it finished roughly where it had closed on Wednesday. For a company whose shares had gained 36 percent in a single day after the same report exactly one year earlier, that is a remarkably sober verdict. It only makes sense once you read the numbers that were not in the headline.
The cash flow statement tells a different story from the income statement
Oracle spent 28.5 billion dollars on capital expenditure in the quarter, almost all of it on data centers, graphics processors, power and cooling. In the same quarter a year earlier the figure was 8.5 billion. In the whole of fiscal 2025, which ended in May 2025, Oracle had invested 21.2 billion dollars. A single quarter of the current year therefore exceeds a full year that was itself a record at the time. Fiscal 2026 came in at 55.7 billion, and for fiscal 2027 chief financial officer Hilary Maxson guided to a range of 90 to 95 billion dollars.
Set that figure beside the revenue guidance to grasp its scale. Oracle expects at least 90 billion dollars of revenue this fiscal year. It also expects 90 to 95 billion dollars of capital expenditure. The company will spend at least as much on building data centers this year as it collects in total, from every database license, every application subscription, every support contract and every piece of hardware. Among the hyperscalers the ratio of capex to revenue runs, roughly, from about a fifth at Amazon and Alphabet to around half at Meta. At Oracle it is about one hundred percent. This is no longer a software company that happens to run a cloud. It is an infrastructure builder that happens to sell software.
Operating cash flow in the quarter was 23.1 billion dollars, up 184 percent. The number sounds impressive, and it was duly highlighted on Thursday. It still was not enough: after capital expenditure, free cash flow came to minus 5.4 billion dollars. The company spent more than it took in, despite record earnings, despite a GAAP operating margin of 35 percent, despite net income of 4.7 billion dollars. At Oracle, profit and cash have come apart.
Eleven billion of it was paid in advance
The detail that makes the numbers legible sits in an aside from the CFO. Hilary Maxson said on the call, verbatim: “Our CapEx for the quarter was 28 billion dollars, leading to negative free cash flow of 5 billion dollars. And our net cash CapEx, so net of prepayments, was 18 billion dollars for the quarter.” The gap between 28 and 18 billion is customer prepayments. Concretely, the 23.1 billion dollars of operating cash flow included 11.4 billion dollars that customers paid for computing capacity they have not yet received.
Strip that prepaid money out and Oracle generated roughly 11.7 billion dollars from its ongoing business in the quarter. Against that stood 28.5 billion dollars of capital expenditure. The gap of nearly 17 billion dollars was closed in two ways: with the 11.4 billion of customer prepayments and with the sale of Oracle’s own shares through an at-the-market program that was completed during the quarter and raised 19.9 billion dollars net. The share count rose 3.1 percent year on year to 3.0 billion as a result. An investor who owned one percent of Oracle a year ago owns 0.97 percent today, and received nothing in return.
That is the core of the quarter, and it fits in one sentence: in the summer of 2026 Oracle built 28.5 billion dollars of data centers, and roughly 40 percent of that was paid for by customers in advance, roughly 40 percent by shareholders through dilution, with the underlying business covering the remainder. That is not an accusation. It is a description of the business model as it currently operates.
Why a prepayment is not the same as revenue
A customer prepayment is cash coming in on the cash flow statement. On the balance sheet it is a liability: Oracle owes the customer computing in return. Revenue is recognised only as the capacity is delivered and consumed, quarter by quarter over the life of the contract. For the customer the prepayment is a reservation: anyone who wants to be sure of getting graphics processors in 2027 and 2028 pays part of the price now, because capacity is scarce. Oracle reported utilisation of its GPU fleet at 97.9 percent and contract renewals signed at an average premium of 20 percent to the original price. In a seller’s market like that, prepayment is rational on both sides.
It has one property to keep in mind, though: it does not repeat. A customer pays up front once, and then Oracle delivers against that money for years. In the quarters when the capacity is actually running, revenue appears but no new cash arrives, because the cash was already there. The 23.1 billion dollars of operating cash flow in the first quarter is therefore not a base you can simply multiply by four. It contains money for deliveries that will happen later, and in that sense it is an advance on future quarters.
Co-chief executive Clay Magouyrk stated the principle openly on the call: “Capital is still required to do this work, but it doesn’t all have to flow from Oracle’s side. It does not have to be Oracle CapEx.” He named three routes: supplier financing, under which vendors get paid as customers pay Oracle; customers placing their own hardware in Oracle data centers; and the prepayments themselves. For the full year, net cash capital expenditure is therefore meant to come in at no more than 70 billion dollars even though gross capex will be 90 to 95 billion. The 20 to 25 billion dollar difference is carried by suppliers and customers. That is the real news of the quarter: Oracle has stopped financing its data centers alone, because it can no longer finance them alone.
125 billion of debt, and rates are going the wrong way
Unlike Microsoft, Alphabet or Amazon, the company has no net cash pile to draw on. Oracle carries roughly 125 billion dollars of debt. Interest expense in the quarter was 1.43 billion dollars, up 55 percent, which annualises to a little over 5.7 billion, or about seventy percent of one quarter’s adjusted operating income of 8.2 billion. For fiscal 2027 Oracle has announced 40 billion dollars of capital raising: the 20 billion from the completed share sale, plus a further 20 billion or so in early 2027, originally planned as bonds but, after conversations with the rating agencies, now possibly again partly in equity.
The timing is poor. The ten-year Treasury yield stood at 4.95 percent on Thursday, the highest since 2023, and futures markets put the probability of a Federal Reserve rate hike next week at 73 percent. Every bond Oracle issues in 2027 will cost more than the ones from 2025, and a company with 125 billion dollars of debt and free cash flow of minus 23.7 billion in its last fiscal year is no longer a first-tier borrower. That is precisely why the rating agencies are steering it toward equity, and precisely why the stock fell in the regular session on Thursday as yields rose: before the results were even out, the market had repriced the cost of financing, not the business.
664 billion of backlog, and how much of it arrives
The 664 billion dollar backlog is the number Oracle uses to justify its spending. It is real, in the sense of signed contracts, and at the same time it is a number that demands care. It rose 209 billion dollars year on year but only 26 billion from the prior quarter, after growing 85 billion in the spring quarter. The pace of accumulation has slowed noticeably, even though Oracle says it signed more than 30 billion dollars of new AI contracts in the quarter. The difference is explained by the fact that every quarter a growing slice of the backlog is now worked off as revenue.
Oracle expects about half of the backlog to convert to revenue within 36 months, roughly 330 billion dollars over three years, or about 110 billion a year. That is more than the entire 90 billion of revenue the company expects for the current year, which shows how heavily the backlog is weighted toward 2028 and 2029. The largest single item is known: OpenAI has committed to buying about 300 billion dollars of computing capacity over five years starting in 2027, and Meta has committed 20 billion. A substantial share of the backlog thus hangs on a single customer whose own business does not yet make a profit and which will pay its Oracle invoices out of funds it has yet to raise.
The 11.4 billion dollars of prepayments cut both ways in this context. They are the strongest argument that the demand is genuine: nobody wires eleven billion for capacity it does not need. They are also a signal that Oracle needs the prepayments to build the capacity in the first place. It is a loop in which the customer lends the supplier the money the supplier uses to build the product the customer later buys. As long as the customer stays solvent, it works. And as long as the capacity ends up 97.9 percent utilised, Oracle earns on it. Both conditions hold today. Neither is guaranteed.
What the numbers mean for American investors
The most direct read-across is to the companies that get paid when Oracle builds. Nvidia sells the more than 300,000 GPUs Oracle deployed in the quarter, nearly triple the prior quarter. Broadcom and Arista supply the networking. The 850 megawatts Oracle brought online in the quarter, 73 percent of everything it added in all of fiscal 2026, is exactly the demand that shows up in the order books of Vertiv for cooling, GE Vernova for gas turbines and Constellation Energy for power. All of these benefit while Oracle spends 90 to 95 billion, and all of them are exposed if the financing that makes that spending possible tightens. Note the mechanism Magouyrk described: supplier financing means some of those vendors are now waiting for Oracle’s customers to pay before they themselves get paid. The supply chain has become a lender.
The second comparison is CoreWeave, which runs the same model in purer form: rented GPUs, contracted backlog, heavy debt, customer concentration in a handful of AI labs. CoreWeave’s bonds trade at yields that price a meaningful probability of distress, and its equity has swung accordingly. Oracle is CoreWeave with a 5.5 billion dollar legacy software business attached, and that business, which shrank 3 percent in the quarter, is what still lets Oracle borrow at investment-grade rates. The 4.2 billion dollar applications segment grew just 10 percent, and co-CEO Mike Sicilia spent his time on the call arguing that AI is “an accelerator, not a replacement for packaged applications.” For Microsoft, Salesforce and Workday holders, that is the same argument they are hearing from their own management, and it is equally unproven everywhere.
For anyone holding the stock directly, Oracle pays a quarterly dividend of 0.50 dollars, about 1.5 billion dollars per quarter in total. Qualified dividends and long-term capital gains are taxed at 0, 15 or 20 percent depending on income, plus the 3.8 percent net investment income tax above the thresholds, and anyone who sold into the 55 percent drawdown from the 2025 peak and wants to buy back should mind the 30-day wash-sale rule. Worth noting: Oracle continues to pay that dividend while free cash flow is negative. Arithmetically, the payout is funded by the proceeds of the share sale. Old shareholders receive the money new shareholders put in.
The counter-argument: why the bulls may still be right
There is a reading in which all of this is not merely acceptable but correct. It runs as follows: Oracle is the only large provider with utilisation near 98 percent and renewal pricing 20 percent above the original contract. Every data center built today is sold before it stands. Under those conditions building as fast as possible is not waste but the only rational strategy, because every month of delay is a month in which the customer goes elsewhere. The 90 to 95 billion dollars of capex is, on this view, not a bet but the fulfilment of contracts already signed. The backlog is the collateral.
Then there is valuation. After falling about 55 percent from its September 2025 high, Oracle trades at roughly twenty times expected earnings while revenue grows 30 percent and infrastructure revenue 121 percent. Few names in the S&P 500 offer that combination of growth and multiple. Analyst targets average about 257 dollars, in a range from 210 at Morgan Stanley to 400 at Guggenheim. Even the most cautious target sits almost 40 percent above the current price. And the guidance raise for the full year, from 27 to 29 percent growth to 34 percent, is no small thing: it means capacity is coming online faster than planned, not slower.
One can take those arguments seriously and still see where they limp. The multiple argument assumes earnings per share actually land at 8.10 dollars, and that is an adjusted figure containing depreciation on data centers whose useful lives Oracle itself sets. The utilisation argument applies to today’s fleet. Whether a fleet three times the size in 2028 will also be 98 percent utilised depends on whether OpenAI and the other large customers can honour their contracts then. A backlog is a promise, not a payment.
Outlook: three numbers for the coming quarters
The next date is Oracle AI World, October 25 to 28 in Las Vegas, with an investor day on October 28. There the company will have to spell out its capital plans beyond 2027, and the question in the room will not be how much gets built but who pays for it. Three numbers decide whether this quarter’s story continues or turns.
First, the share of prepayments in operating cash flow. In the first quarter it was 11.4 of 23.1 billion, nearly half. If that share keeps rising, financing risk shifts ever further onto customers and the cash flow statement says ever less about the ongoing business. If it falls because customers will not or cannot keep prepaying, Oracle has to close the gap in the capital markets at terms that are deteriorating right now. Second, the quarterly change in backlog. From 85 billion in the spring to 26 billion in the summer is a sharp deceleration; if the figure falls below infrastructure revenue in the second quarter, the backlog shrinks for the first time. Third, the terms of the early-2027 capital raise. If it comes as a bond at a spread the market reads as a warning, or as another share sale that pushes dilution past six percent in two years, that will answer the question the market asked in Thursday’s regular session.
Oracle delivered a record quarter, and at the same time it showed that it could not pay for that record quarter on its own. Both things are true. The mistake would be to see only one of them.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
