Larry Ellison reversed a $7.5 billion Oracle stock sale a day after it went public. The real story is Oracle's negative free cash flow.

Oracle told investors on Saturday, September 12, 2026 that Larry Ellison had canceled a 10b5-1 trading plan to sell up to 50 million shares, worth roughly $7.5 billion when the plan surfaced. He sold none of it, and Oracle says he has no other plans to sell. The reversal came one day after the plan itself became public knowledge and rattled a stock that was already down sharply this year on worries about how Oracle is paying for its AI data center buildout.

The 36 hours nobody planned for

The plan was adopted on June 22, 2026, under Rule 10b5-1, the mechanism executives use to schedule stock sales in advance so they can't be accused of trading on inside information. It was set to run until October 24. A regulatory filing disclosed the plan's existence in mid-September, and by the next day Oracle had put out a two-sentence statement killing it: "No Oracle stock was sold under that plan, and he has no other plans to sell any of his Oracle stock." No explanation given, which is its own kind of statement from a company that usually explains everything.

The timing tells you why it mattered. Oracle's stock had fallen roughly 16 to 18 percent since the plan was adopted in late June, according to reporting on the filing, and TechCrunch noted the stock was down 22 percent since the start of the year at the time of the cancellation. Zoom out further and the drop is worse: shares touched a 52-week high of $345.72 in September 2025 and have traded more than half below that peak at various points this year. A sale that looked routine in June, 50 million shares at roughly $175 apiece, looked like the founder cashing out into weakness by September. Canceling it removed that read, but it didn't remove the reason the read existed in the first place.

What's actually stressing the balance sheet

Here's the part that doesn't show up in a headline about one man's stock plan. Oracle closed fiscal year 2026 (ended May 31) with capital expenditures of $55.7 billion, up 162 percent year over year, funding the AI data centers it's racing to build. Free cash flow for that year came in at negative $23.7 billion. Operating cash flow was strong, a record $32 billion, but the capex outran it by a wide margin.

The quarter after that didn't ease up. Oracle's own Q1 FY2027 earnings release, published September 10, just two days before the Ellison story broke, reported free cash flow of negative $5 billion for the quarter and record operating cash flow of $23 billion, up 184 percent. To help fund the gap, Oracle completed a $20 billion stock sale through an at-the-market equity program during the quarter, part of a financing plan that already included $43 billion in new debt and $5 billion in equity raised during fiscal 2026. That debt load has made Oracle the second-largest non-financial issuer in the Bloomberg US Corporate Bond Index, behind only Amazon, with something like $117 billion outstanding. S&P Global Ratings cut Oracle's long-term credit rating a notch, to BBB-, in July, one step above speculative grade, and Moody's has put a negative outlook on its own rating.

Against all of that, Oracle keeps posting a genuinely enormous backlog. Remaining performance obligations, the contracted AI cloud revenue Oracle expects to recognize over time, hit $664 billion at the end of Q1 FY2027, up $209 billion year over year. Oracle booked more than $30 billion of new AI cloud contracts in that quarter alone. By most reporting, OpenAI accounts for roughly half of that backlog on its own.

Ellison's own money is tangled in this too

Ellison's family trust owns about 40.6 percent of Oracle, something like 1.16 billion shares. According to Oracle's own proxy filing, as reported by the Motley Fool and by IFR, 346 million of those shares, close to 30 percent of his position, are already pledged as collateral for personal loans. Oracle's board has said the loans fund outside business ventures rather than margin accounts and that it believes Ellison can repay them without selling stock, but the company also carved out a specific exception to its own no-pledging policy just for him.

Separately, and this part deserves a hedge because it's still working through the courts, Ellison gave an irrevocable personal guarantee last December covering $40.4 billion of the equity financing behind Paramount Skydance's acquisition of Warner Bros. Discovery. Twelve state attorneys general sued this year to block that merger, a judge froze the deal while the case proceeds, and Paramount has agreed it won't close before June 1, 2027. None of that touches Oracle's operations directly. But it means a meaningful share of the man who controls 40 percent of the company has personal financial obligations that get more expensive, relative to his wealth, every time the stock falls.

Why this matters even if you never touch Oracle Cloud

None of this is a reason to panic if you build on AWS, or a different neocloud entirely, or your own hardware. But it's a good prompt to actually check something most of us never look at: whose infrastructure is underneath your AI vendor. Oracle isn't only an AI company spending heavily on its own products; it's increasingly a landlord, renting compute capacity to other companies and labs that then resell access to you. A $664 billion backlog is real demand and it's also, by definition, a promise of future revenue rather than proof that today's pricing or capacity commitments hold up if the financing gets harder or a major tenant renegotiates.

If a tool you depend on runs on Oracle Cloud Infrastructure, or on any provider carrying this much debt against negative free cash flow, that's not a reason to leave tonight. It's a reason to spend fifteen minutes finding out. Check the vendor's status page or documentation for which cloud they run on. If they don't say, ask them directly, most will answer. If the answer is a single highly leveraged provider and you have no fallback, that's the actual risk this story points at, not whether Larry Ellison sells or doesn't sell his own stock.

What I'd actually do

I'd treat the canceled sale as noise and the balance sheet as signal. Concretely: for any AI API or hosting service central to what I'm building, I'd find out which cloud it runs on and whether the vendor has a documented multi-cloud or multi-region fallback, not because Oracle is about to collapse, but because concentration risk anywhere in your stack is worth pricing in before it's forced on you during an outage or a pricing change. I keep at least a tested alternate path for anything my business actually depends on, and this is exactly the kind of story that should push that habit rather than sit as a headline you scroll past.

Where I could be wrong: the record backlog and the completed $20 billion equity raise suggest Oracle's lenders and biggest customers are still betting the financing works out, and insiders reversing a planned sale is often read as a confidence signal rather than a warning one. Negative free cash flow funding real, contracted demand isn't automatically a crisis; plenty of infrastructure buildouts look like this before they pay off. I'm weighting the leverage more heavily than the market currently is, and the market could simply be right.

Author

Lukas

@lukcombinator

Sources

• Larry Ellison Cancels His Plan to Sell Oracle Stock
• Larry Ellison cancels $7.5 billion sale of Oracle stock
• Oracle Announces Q1 Results Driven by Triple Digit Growth in Cloud Infrastructure Revenues
• Larry Ellison Owns 40.6% of Oracle and Has Backstopped $40.4 Billion of a Media Deal
• Oracle stock shrugs off S&P downgrade to 'BBB-', but $160B debt shadow looms