Oracle Booked $664 Billion. Its Free Cash Flow Is Still Negative.

Oracle reported its fiscal first quarter after the close on September 10, and the number everyone grabbed was $664 billion. That is the company's remaining performance obligations, the accounting term for revenue customers have contracted to pay but that Oracle has not yet delivered. Put it next to the company's own guidance of at least $90 billion in revenue for the full fiscal year and the ratio looks almost absurd. More than seven years of sales, already signed. The stock rose about 6 percent the following morning. Here is the part that did not make the headline: Oracle has now posted negative free cash flow for five consecutive quarters, carries more than $100 billion of debt, and was downgraded this year by S&P to BBB minus, a single notch above speculative grade. Both pictures are accurate. Reconciling them is the most valuable half hour a self-directed investor can spend on the AI trade right now.

The quarter was not a mirage

Start with what actually happened, because the operating results were strong by any standard. Total revenue came in at $19.3 billion, up 30 percent from a year earlier. Cloud infrastructure revenue, the piece that rents computing power to companies training and running AI models, rose 121 percent to $7.4 billion. That was the ninth consecutive quarter in which cloud infrastructure growth accelerated rather than merely continued, which is rarer than it sounds. Total cloud revenue climbed 62 percent to $11.6 billion, while the older applications business grew a more pedestrian 10 percent to $4.2 billion. Net income rose roughly 60 percent to $4.7 billion. Adjusted earnings of $1.92 a share beat the roughly $1.73 analysts had penciled in. Management raised full-year guidance to at least $90 billion in revenue and $8.10 in adjusted earnings per share.

The physical build behind those numbers is just as striking. Oracle said it brought roughly 850 megawatts of additional data center capacity online during the quarter and delivered more than 300,000 GPUs to customers. GPU utilization ran at 97.9 percent. There is no demand problem here. Every chip Oracle can plug in has someone waiting to rent it.

So why has the stock spent most of 2026 going the wrong way? Coming into this report it was down roughly 17 percent on the year and more than 50 percent below the peak near $331 it touched in September 2025. The answer is not on the income statement.

Backlog is a promise. Cash is a fact.

Remaining performance obligations is a real accounting disclosure with real rules behind it, not a marketing figure. But it measures what customers have agreed to pay, not what they have paid, and it says nothing about what it will cost Oracle to fulfill the agreement. A backlog number tells you about demand. It tells you almost nothing about the capital required to convert that demand into delivered service, or about who absorbs the loss if the customer cannot pay.

That second question is where Oracle gets uncomfortable. Bank of America has estimated that OpenAI accounts for more than half of Oracle's contracted backlog. Sit with that for a second. The single largest component of Oracle's forward-looking disclosure is a commitment from a private company that is itself funding operations through enormous outside capital raises rather than profits. Oracle has said publicly that it is confident in OpenAI's ability to raise the money and meet its obligations. That confidence may well be warranted. But anyone buying Oracle on the strength of the backlog is, whether or not they realize it, taking a sizable position in OpenAI's future fundraising.

The financing stack is the actual story

Here is the arithmetic the credit market has been staring at all year. Oracle's capital spending more than doubled in fiscal 2026 to roughly $55.6 billion. Operating cash flow grew 54 percent to about $32 billion, which sounds terrific right up until you subtract the capex. Free cash flow for the fiscal year landed at negative $23.7 billion.

That gap has to be filled from somewhere, and Oracle filled it by borrowing. The company raised roughly $43 billion in debt and about $5 billion in equity during fiscal 2026, and has signaled plans to raise another $45 billion to $50 billion in calendar 2026 through some combination of the two. Set against existing debt north of $100 billion, that is a balance sheet being rebuilt around a single bet.

The bond market noticed before the equity market did. Oracle's credit default swap spreads, essentially the cost of insuring against a default, climbed to record levels by early August. S&P cut the rating to BBB minus with a negative outlook. TD Cowen has estimated that fulfilling existing agreements will require Oracle to procure on the order of three million GPUs plus supporting equipment, and reported that a number of U.S. banks have pulled back from lending against Oracle-linked data center projects, with borrowing costs rising where lenders remain willing. In December 2025, Blue Owl, a firm that had been a significant Oracle partner, declined to fund a $10 billion data center project in Michigan.

None of this means Oracle is in trouble. It means the company has chosen a materially different funding model from its largest competitors. Amazon, Alphabet, Microsoft and Meta are spending staggering sums on AI infrastructure too, with the five largest U.S. technology buyers collectively on track for somewhere around $660 billion to $690 billion of capital expenditure in 2026, up from roughly $380 billion in 2025. Dell'Oro Group expects global data center capex to top $1 trillion this year. Goldman Sachs has projected hyperscaler capex of about $1.15 trillion across 2025 through 2027, more than double the $477 billion spent from 2022 through 2024. The difference is that the others are largely funding it out of operating cash flow thrown off by mature, high-margin businesses. Oracle is funding it out of the bond market.

The depreciation clock nobody wants to discuss

There is a second risk buried beneath the first, and it has nothing to do with credit. AI chips are improving faster than data centers can be built. A facility conceived around one generation of accelerator can be commercially stale relative to customer preference before the power is even energized. That is not hypothetical. OpenAI has reportedly decided against expanding its arrangement with Oracle at the Abilene, Texas site because it wants clusters built on newer Nvidia silicon than the Blackwell generation that site was designed around.

Investors are used to thinking about data centers as long-lived infrastructure, closer to a toll road than to a laptop. If the useful economic life of the expensive contents is really three or four years rather than six, then the depreciation schedules embedded in nearly every AI infrastructure model on Wall Street are too generous, and reported earnings across the sector are flattered accordingly. Oracle is not uniquely exposed to that math. It is simply the most leveraged to it.

What to actually watch

Oracle deserves credit for adapting. The company indicated that its newest contracts were signed with prepayment terms or bring-your-own-hardware arrangements that do not require incremental capital from Oracle. That is a meaningful structural shift, and if it becomes the norm rather than the exception, the risk profile changes considerably.

For anyone trying to read the AI buildout rather than trade it, Oracle has become the sector's stress test. Watch whether free cash flow turns, and on what timeline management says it will. Watch the composition of new bookings, specifically how much of it comes from customers who are themselves profitable. Watch the credit default swap spreads, which have been a better leading indicator on this name than the share price. And watch whether those prepaid, customer-financed contract structures spread to the rest of the industry or stay a one-off.

The demand is real. That was never seriously in doubt, and the 97.9 percent utilization figure settles it. The question the market is asking is narrower and harder. Who funds the gap between signing a contract and collecting on it, and what happens to them if the timeline slips?

This article is written for educational and informational purposes only and does not constitute financial or legal advice. The views and analytical frameworks presented draw on publicly available information and reported commentary from industry participants. Readers are encouraged to consult primary sources and form their own informed views on these complex topics.