Oracle shares surged roughly 85% last September when the market woke up to the size of its AI cloud backlog. The stock hit a 52-week high of $345.72. Since then it has round-tripped hard, falling about 58% from that peak, including a 19% drop in a single week in late June — Oracle’s worst week since the 2001 dot-com crash.

Here’s the confusing part. The demand problem you’d expect behind a crash like that doesn’t exist. Oracle just posted record results, beat estimates, and raised guidance for the year ahead. The stock isn’t falling because customers are leaving. It’s falling because of how Oracle is paying for its own growth.

The boom is real

Full-year revenue hit $67.4 billion, up 17%. Cloud revenue reached $34.0 billion, up 39%. Oracle Cloud Infrastructure, which rents out compute and GPU capacity for AI workloads, surged 77% to $18.1 billion. Non-GAAP earnings per share grew 27%. The standout number: Oracle’s backlog of signed-but-undelivered contracts more than quadrupled to $638 billion — the basis for management’s confidence in next year’s guidance of $90 billion in revenue. The demand is already under contract, not just hoped for.

The financing problem

Oracle is funding a massive AI data-center buildout largely with debt and new stock, against deeply negative cash flow. Capital spending rose 162% to $55.7 billion this past year. Free cash flow came in at negative $23.7 billion. S&P just cut Oracle’s credit rating to BBB-, one step above junk, citing exactly this dynamic.

The shift in strategy is stark. Oracle spent over a decade buying back its own stock, shrinking its share count by 45%. That’s essentially stopped — buybacks fell about 80% this year. Instead, Oracle raised $5 billion through preferred stock in February, plans a further $20 billion equity raise, and expects to raise tens of billions more in debt and equity over the next three years. Total debt already sits near $130 billion, with subsequent issuance pushing that closer to $167 billion.

One customer, an outsized share of the risk

A huge slice of that backlog traces to one not-yet-profitable customer: OpenAI. The deal is a $300 billion, five-year cloud contract requiring Oracle to deploy roughly $135 billion of hardware, with about half of that spend due before matching revenue shows up. Reporting attributes roughly $300 billion of the entire $638 billion backlog to this single relationship.

Moody’s compares the arrangement to one of the world’s largest project financings, minus the protective structure that usually comes with one. Oracle’s own corporate balance sheet — not a walled-off vehicle — is absorbing the counterparty risk directly. It’s not theoretical: banks have reportedly grown cautious about the exposure, and at least one developer redirected a data-center project from Oracle to Microsoft over the same concern.

A real moat, in a race Oracle isn’t leading

Oracle’s legacy database and ERP business has a genuine, durable moat — deep enterprise lock-in that funds high-margin recurring revenue. OCI is different: Oracle is essentially renting out Nvidia GPUs, competing on price and speed against AWS, Microsoft, and Google, all with larger, more diversified balance sheets. Its smartest counter-move is a multicloud database strategy: placing Oracle Database inside rivals’ data centers, turning them into distribution partners.

Leadership changed hands last September. Longtime CEO Safra Catz moved to Executive Vice Chair after selling over $2.5 billion in stock; co-CEOs Clay Magouyrk and Mike Sicilia now run the company. Larry Ellison remains Chairman and CTO and, holding roughly 40% of shares, is still the largest shareholder. A pending bondholder lawsuit alleges Oracle understated its likely future borrowing in a September note offering — unresolved, but worth tracking.

Cheap, or cheap for a reason?

On paper, Oracle looks inexpensive: about 17–18x forward earnings, with a PEG ratio near 0.6 relative to its growth rate. Average analyst targets cluster around $225–270, well above today’s price — though those targets are being revised down, and several predate the credit downgrade. Bond markets tell a different story, actively pricing in rising default-adjacent risk. Both things can be true: Oracle may be undervalued if the buildout pays off as contracted, and fairly priced once you weigh the real chance it doesn’t.

Where this lands

This isn’t a Buy — the balance sheet stress is confirmed by an independent rating agency, tied to risk concentrated in a customer whose own ability to fund its side of the deal is unproven. It isn’t an Avoid either — demand is real and contracted, guidance was reaffirmed, and the stock has gotten genuinely cheaper. The right move is watching: a free cash flow inflection, credit stabilization, or signs the OpenAI concentration is diversifying would all shift the calculus.

Read the complete Oracle (ORCL) research report — full financials, valuation model, and competitive breakdown — on our Reports page.