Oracle's AI cloud backlog hit $664 billion in its fiscal 2027 first quarter, reported September 10, 2026, up $209 billion from a year earlier and $26 billion just since the prior quarter. That headline landed the same week Oracle's stock actually fell despite beating Wall Street's revenue and earnings estimates, and it landed two months after S&P Global cut Oracle's credit rating to BBB-, the lowest rung of investment grade, over the debt and negative cash flow the AI buildout is running on. The backlog is real. So is the bill for funding it, and that second half is the part most hype-driven coverage of this number leaves out.
What Did Oracle Actually Report on September 10?
Oracle beat estimates across the board. Revenue came in at $19.35 billion, up about 30% year over year and above the roughly $19.1 billion Wall Street was expecting, with adjusted earnings per share of $1.92 against a $1.74 consensus. Cloud infrastructure revenue, the part of the business actually renting out AI computing power, surged 121% year over year to $7.4 billion, and Oracle said GPU utilization across its data centers is running at 97.9%, meaning it is not sitting on idle capacity waiting for customers. Remaining performance obligations, Oracle's name for its confirmed order backlog, reached $664 billion, beating the $630.6 billion analysts had modeled.
Why Did the Stock Fall on a Beat?
Oracle's stock dropped roughly 7% in the days around the report, even with every headline number coming in ahead of estimates, because investors are increasingly pricing the cost of funding that backlog rather than just its size. Oracle spent $28.5 billion on capital expenditures in this single quarter alone, and management pointed to peak annual capex of $90 billion to $95 billion still ahead as the AI buildout continues. A backlog that big is only as good as Oracle's ability to keep building the data centers to fill it, and that construction bill is what the market reacted to, not the order book.
Where Is the $664 Billion Actually Coming From?
More than $30 billion in new AI cloud contracts were signed in this quarter alone, and Oracle says its non-OpenAI backlog has more than doubled over the past year, which matters because a year ago the bear case against Oracle's AI story was that it was really just one customer's bet. A broader customer base cuts against that concentration risk directly. Just as important structurally, the majority of this quarter's new contracts came through prepayment or "bring your own hardware" arrangements, meaning a meaningful share of this backlog does not require Oracle to put up the capital itself before it gets paid. That's a materially different risk than a backlog funded entirely off Oracle's own balance sheet, even though both show up in the same $664 billion headline.
What's the Real Risk Here? The Part the Backlog Number Leaves Out
The risk sitting underneath this story is Oracle's own balance sheet, not whether AI demand is real. S&P downgraded Oracle's long-term credit rating to BBB- in July 2026, citing mounting debt and sharply negative cash flow, and BBB- is the last stop before a company's debt is considered speculative-grade, or "junk." Moody's has attached a negative outlook of its own on top of that. For the fiscal year that ended May 31, 2026, Oracle spent more than $55 billion on capex and its free cash flow turned negative by almost $24 billion. Oracle now carries roughly $117 billion of debt outstanding, ranking second only to Amazon among non-financial issuers in the Bloomberg US Corporate Bond Index. None of that means the demand behind the $664 billion isn't genuine. It means Oracle, a company with decades of history running a conservative balance sheet, is now more financially leveraged to that demand converting into actual cash, on schedule, than it has arguably ever been.
How Does This Compare to Nvidia's Backlog Story?
Both stories start from the same place: real, confirmed, dollar-verified demand for AI infrastructure that isn't a forecast or a hopeful slide deck. Where they diverge is what stands between that backlog and the revenue it eventually becomes. Our breakdown of Nvidia's $1 trillion backlog and the power-grid backlash coming for it covers a bottleneck entirely outside Nvidia's control: whether local utilities can physically deliver enough electricity to plug in the chips that are already sold. Oracle's bottleneck is closer to home. It's Oracle's own credit rating, debt load, and ability to keep raising capital cheaply enough to keep building, which makes this as much a corporate-finance story as it is an AI story.
What to Actually Do With This
- Separate backlog size from backlog funding risk. "Is the demand real?" and "Can the company afford to build fast enough to capture it without straining its balance sheet?" are two different questions, and Oracle's September report answered them very differently.
- Watch credit rating actions and bond spreads, not just the RPO headline. A downgrade toward junk status would raise Oracle's borrowing costs at exactly the moment it needs to keep spending tens of billions a year, a feedback loop that doesn't show up in a backlog number.
- Don't treat one company's execution as a bet on "AI" broadly. A broad index fund already owns a slice of the hyperscalers, the chipmakers, and the infrastructure providers racing to build this out, without requiring you to correctly call how any single company's balance sheet holds up.
The backlog number is the part of this story that gets the headline. The credit downgrade, the negative free cash flow, and the $90 billion-plus in capex still ahead are the part that determines whether that backlog ever fully shows up as cash. Both halves are true at the same time, and that tension is exactly why this is a story worth following past the number everyone's quoting.
—
Spicy Investing