Oracle's Cloud Just Hit a Record. The Stock's Problem Is Cash, Not Demand

Oracle turned in exactly the kind of cloud quarter the AI trade wanted. Cloud revenue rose 62% to a record $11.6 billion in the fiscal first quarter, infrastructure more than doubled, and management raised the full-year forecast. The shares still fell.

That disconnect is the story — not because the numbers were weak. They weren't. Because the market was never really asking whether Oracle can sell AI capacity. It was asking whether Oracle can turn that capacity into cash before the buildout burns through the balance sheet.

The demand half looks as close to perfect as it gets

Run the scoreboard on the quarter ended August 31, and Oracle beat on both lines. Total revenue came in at $19.35 billion, up 30%, above the roughly $19.1 billion Wall Street expected, and adjusted earnings per share of $1.92 cleared the ~$1.74 consensus. The headline number, though, is the 62% jump in cloud revenue to a record $11.6 billion — now 60% of everything Oracle sells.

• Cloud revenue: $11.6 billion, up 62%, 60% of total revenue.
• Cloud infrastructure (the AI category): , up 121%.
• New AI cloud contract bookings in the quarter: more than $30 billion.
• Backlog (RPO): a record , up $209 billion year over year.

That last line deserves a translation. Remaining performance obligations are revenue Oracle has contracted but not yet counted, because the service hasn't been delivered. A $664 billion backlog means the demand is real, it is written down, and the buyers are named — Oracle lists OpenAI, Nvidia, Uber, Amazon and Microsoft among its AI cloud customers.

Management used that visibility to raise guidance: full-year fiscal 2027 revenue to at least $90 billion, with cloud growth guided to accelerate to roughly 64–70% next quarter. By every bookings measure, this was a strong quarter.

The other half is where the market is actually looking

So why did the pop in extended trading — shares were up about 7% right after the release — fade into a down session, with the stock trading around $153, roughly half its 52-week high and down about 22% on the year?

The answer is the cash-conversion half of the AI commercialization chain. Capital spending accumulates first, revenue follows, and free cash flow repairs last. Oracle built record infrastructure in the quarter — delivering 850 megawatts of new data-center capacity and more than 300,000 GPUs — and the bill came due all at once:

• Capital expenditures in the quarter: , versus $8.5 billion a year earlier.
• Free cash flow: negative about $5 billion in the quarter, versus a $0.36 billion shortfall a year ago.
• Operating cash flow hit a record $23 billion, up 184% — and the buildout still consumed it and more.

To fund the program, Oracle sold about $20 billion of new stock through an at-the-market equity program in the quarter, and it has said it will raise more capital during the year. The full-year capital plan is roughly $70 billion, plus another $20–25 billion in prepayments for components.

Now the tension, in one number: that $664 billion backlog converts to revenue at only about 50% over the next 36 months. So the biggest backlog in the industry does not arrive as cash for years, while the capex bill is due now. This is why negative free cash flow is the live question, not a footnote: last fiscal year Oracle's free cash flow was already roughly negative $24 billion, and the pattern is continuing.

What the quarter actually settled — and what it didn't

Here is the cleanest way to read it. The quarter settled the demand question: AI infrastructure demand is real, enormous, financed and contracted. The market stopped doubting that months ago. What it has not stopped doubting is the second stage — whether Oracle can convert that backlog into recognized revenue fast enough to close a deficit that analysts have projected to stretch deeper this year, all while carrying a heavy debt load and concentrating a large share of its business around a handful of AI customers.

That is the rerating condition. More bookings do not rerate the stock anymore; cash conversion does. Oracle rerates when the backlog starts producing free cash flow — when the deficit shrinks and turns positive — not when management books another $30 billion of contracts.

The way to keep score from here is to watch two things each quarter. First, whether new AI bookings keep coming in at the >$30 billion clip, because that tells you demand is holding. Second, whether the free-cash-flow deficit narrows as the newly installed capacity starts generating cash rather than absorbing it.

The signal that would break the thesis is a slowdown on either side. If new AI bookings decelerate, the demand story itself is being tested, not just the financing. If Oracle stretches out its data-center buildout — management insisted there were no delays at sites such as its New Mexico project — that would read as a loss of conviction in the conversion, and the stock would retest its lows. For now, Oracle is a bet on the second half of the chain: not whether it can sell the capacity, but whether it can get paid for it fast enough.