NEW YORK

Larry Ellison has initiated a Rule 10b5-1 trading plan to sell up to 50 million Oracle shares, valued at $7.5 billion at current prices. The plan, adopted June 22, runs through Oct. 24, 2026.

This is a red flag. Ellison, 82 and Oracle's founder, has not sold more than 25,000 shares in a single transaction since 2000, according to FactSet. The timing matters: Oracle shares have collapsed 64 percent from their September 2025 peak above $345.

The sell-off is not mysterious. Oracle's AI infrastructure bet is cannibalizing its balance sheet. The company spent $55.66 billion on capex in fiscal 2026—exceeding its $50 billion guidance—and has guided for $90 billion to $95 billion in fiscal 2027. It burned through $28.5 billion in capex in Q1 alone. S&P downgraded Oracle's credit rating to one notch above junk. The company now carries $117 billion in outstanding debt.

Yet Wall Street is still pretending the strategy works. Oracle beat earnings on a 121 percent surge in cloud infrastructure revenue year-over-year. Remaining performance obligations stood at $664 billion after Q1, up from $638 billion at fiscal year-end.

The problem is concentration risk. Half of Oracle's $638 billion backlog is tied to OpenAI—a single customer described as unprofitable. That tail wags the dog. If OpenAI's funding stalls or its spending plans shift, Oracle's capex justification evaporates.

Institutional investors are pulling back. Brent Thill, analyst at Jefferies, said investors remain concerned about Oracle's balance sheet and aggressive capital deployment in AI cloud.

Ellison's sale is the most direct signal yet that even the founder questions whether Oracle can service $117 billion in debt while spending $90 billion-plus annually on data centers betting on a single customer's runway.