Oracle: The AI Backlog Finally Starts Paying for Itself

Oracle just reported its best quarter in years, and the stock finally moved with it. Fiscal Q1 revenue hit a record $19.3 billion, up 30% from a year earlier, non-GAAP earnings rose 30% to $1.92 a share, and shares jumped about 7% in after-hours trading. None of that is the surprise. The surprise is that a company growing this fast traded down roughly 28% over the past twelve months heading into the print. The market stopped valuing the income statement and started valuing the checkbook.

Why the stock fell while the business accelerated

Oracle's problem was never demand. Its problem was that building the AI data centers to serve that demand consumed cash faster than the income statement generated it, and investors had to finance the gap.

In Q1, capital spending totaled about $28.5 billion in a single quarter — enough to drag free cash flow to negative $5 billion even as operating cash flow hit a record $23 billion. Look further out and the scale is the point: Oracle guided fiscal 2027 capital spending to roughly $90 billion to $95 billion. To fund that, it sold $20 billion of stock this quarter (roughly 5% dilution at recent prices), plans to raise another $40 billion through debt and equity, and had its credit rating cut by S&P to BBB-, one notch above junk. In June, when Oracle guided fiscal 2027 capex above Wall Street expectations and said it would add more debt, the shares fell. A beat in the prior quarter had already earned the stock a ~9% drop, not a rally.

That is the frame the market has been using: a once-staid software company behaving like a capital-hungry AI infrastructure builder, funding the buildout with equity, debt, and a dimmer credit rating.

The quarter that changed the funding math

What investors heard on Thursday's call is that the model is being re-engineered so the backlog pays for itself — and the numbers support it.

The demand side is unequivocally real. Cloud infrastructure revenue more than doubled to $7.4 billion, up 121%; total cloud revenue rose 62% to $11.6 billion. Remaining performance obligations — contracted future revenue not yet on the books — hit $664 billion, up $209 billion from a year ago. GPU utilization is running at 97.9%, and Oracle booked more than $30 billion in new AI contracts in the quarter alone. There is no narrative here; the orders are converting to revenue, and capacity is selling out at a 20% premium to prior contracts on renewals.

The funding part is the more interesting shift. Oracle says the vast majority of its new backlog comes from customers who prepay or bring their own hardware, so those deals don't require incremental capital from Oracle. That is why net cash capital spending — capex net of customer prepayments — came in at only $18 billion against the headline $28.5 billion. Customers are effectively financing their own builds. Management summed up the implication: the RPO growth is now largely self-funding, with no incremental impact on plans to raise capital.

This matters because it attacks the exact fear that drove the year-long slide. If a growing share of new AI contracts carries its own funding, then the mega-capex that produced negative free cash flow and the downgrade becomes less of an open-ended cash burn and more of a finite buildout that customers are underwriting.

The honest bear fact: one customer is the whole story

None of this makes the stock cheap enough to ignore without naming the risk, and the risk is concentration. The vast majority of the backlog's size traces back to OpenAI; Bank of America has estimated OpenAI accounts for more than half of it, and investor Steve Eisman has pointed to roughly $300 billion of exposure tied to that single contract. Oracle is only ~12% of its backlog through this fiscal year, with about a third arriving in the years after, so a large part of the $664 billion lives far in the future.

That is the strongest bear fact against the thesis. All the operating momentum — 30% revenue growth, 121% infrastructure growth, near-full GPU utilization — is real, but an outsized slice of the promised future revenue rests on one customer's willingness and ability to keep paying for compute at this scale. If OpenAI slows, the backlog number, and the valuation story built on it, loses its foundation.

The other tension is the sticker. At roughly 20 times forward earnings against fiscal 2027's guided $8.10 in non-GAAP EPS, Oracle is neither the expensive AI story of its peak nor obviously cheap for a company with this capex and balance-sheet burden.

What would change the call

The stock's late rally marks the first time in a year that a strong quarter produced a positive response, and the mechanism behind it is genuine: demand is converting, pricing power is visible, and the funding for new capacity is shifting onto customers' balance sheets. That combination — accelerating revenue paired with a self-funding backlog — is the case for treating the slide as a reset rather than a falling knife.

It holds only if two things stay true. First, the prepay and bring-your-own-hardware mix has to keep rising, turning the negative free cash flow into a bounded construction phase rather than a structural drain. Second, OpenAI has to keep consuming at the contracted scale. If utilization stays near full and the self-funding share of the backlog keeps growing, Oracle is a fast-growth business available at a reset multiple. If OpenAI's appetite fades, the ~$300 billion concentration means the backlog — the number the whole re-rating depends on — is not as solid as the income statement makes it look. The next proof point is Q2 in about three months, and the quarterly disclosures of how much of the new backlog is prepaid versus company-funded.