Oracle's $664 Billion Backlog Is Finally Landing — and It Carries a Cost

Oracle just reported $11.6 billion of cloud revenue for its fiscal first quarter — up 62% from a year ago, with the infrastructure unit that powers its AI business growing 121%. On the surface, that is about as clean a beat as a cloud company can deliver. The stock, though, has been cut roughly in half over the past year even as revenue accelerates, and the reason is sitting in the same report: converting all that contracted demand into earnings is costing more money than the business is generating.

The figure worth pausing on is not the $11.6 billion itself. It is the source. Oracle's cloud is split between infrastructure (IaaS — the data centers, servers, and GPUs that clients rent) and applications (SaaS — the software on top). The infrastructure side grew 121% to $7.4 billion, while software applications grew a modest 10% to $4.2 billion. That gap is the whole story in miniature: this quarter is being carried by the commodity-scale build-out, not by Oracle's classic software franchise.

The backlog is finally showing up in revenue

To understand why that matters, you need Oracle's most-watched number: remaining performance obligations, or RPO. Think of it as the value of contracts already signed but not yet billed — a backlog of future revenue that Oracle has locked in. That backlog grew another $209 billion over the past year to $664 billion, with more than $30 billion of new AI cloud contracts booked in the quarter alone. The implication investors have churned over for three years is finally showing up in the income statement: the backlog is converting, and at eye-popping rates.

Inside that conversion is the clearest evidence of real, not fictional, demand. Oracle said it delivered more than 300,000 GPUs to AI cloud customers in a single quarter — roughly 73% of all the capacity it delivered in the entire prior fiscal year. Its GPU utilization ran at 97.9%, and GPUs coming up for renewal were renewed or resold at a 20% premium to the prior contract.

That last detail is the one I'd ask you to hold onto. In the inference stage of the AI cycle — where models are already trained and are now being run repeatedly to answer queries — demand is measured by utilization and pricing, not by how many chips a vendor can claim to have sold. At 97.9% utilization and a 20% renewal premium, Oracle is showing pricing power at the most contested, most price-sensitive part of the cycle. That is exactly the signal that separates a backlog that is genuinely durable from one that is defensive inventory.

Every dollar of that growth is being paid for upfront

Here is the other half of the signal. To deliver 300,000 GPUs in one quarter, Oracle spent $28.5 billion of capital expenditures in those three months — up from $8.5 billion a year earlier. Even after customer prepayments, the net cash outlay was roughly $18 billion. Operating cash flow hit a record $23.1 billion, up 184%, but it was not nearly enough: free cash flow for the quarter was negative $5.4 billion, and over the trailing twelve months Oracle has burned nearly $24 billion.

Oracle is funding this build-out two ways, and one of them should matter to any owner. It borrowed — net debt now stands near $98 billion, against a market value of around $440 billion — and, more notably for shareholders, it issued $20 billion of new common stock during the quarter through an at-the-market equity program. Selling new shares to pay for growth dilutes existing holders; the company's own math assumes they do not care because the TAM is growing far faster than the share count. That is a reasonable bet for a monopolist-like position in a huge expanding market, and it is also a real cost that shows up in future per-share earnings.

The market has spent 2026 punishing precisely this trade. The stock fell roughly 59% from its September peak, and the shares were down another 5% on the day of the report even before the earnings landed, after earlier this year handing investors their worst week since 2001 on exactly this cash-burn story. The conventional worry is that Oracle's enormous capex and debt are a spending war it cannot win against Amazon, Microsoft, and Google. That worry is not about whether cloud demand is real — the numbers above say it is. It is about whether the returns on all that capital arrive.

What the conversion has to do to earn back the capital

So the investment question has shifted. It is no longer "is Oracle's cloud growing?" — that is answered. It is whether the backlog converts into cash faster than Oracle is burning it. Management's own plan sketches the trajectory: revenue growing at a 30%-plus clip through the December quarter, cloud revenue up 65% to 71%, and full-year fiscal 2027 revenue of at least $90 billion against a $664 billion backlog. The forward multiple embedded in the stock already assumes a lot of that conversion succeeds.

The variables I would watch are operating, not analyst targets. Does GPU utilization and renewal pricing stay as strong as it is today, or does the flood of new capacity from every hyperscaler push prices back toward cost? Does the $664 billion backlog turn into operating cash flow at a rate that eventually covers the capex, turning that negative free cash flow positive? And does the equity dilution stop, or keep compounding?

Oracle has done the hard part — it turned a three-year promise into a revenue line that is finally growing at triple digits. The next chapter is whether all that contracted demand pays for itself before the capital consuming it becomes the story that wins. For a stock trading at roughly half its high, that is the difference between a beaten-down compounder and a value trap — and the evidence will land in the cash flow statement, not in the growth headline.