The longest bonds in Oracle’s latest sale will pay 8.650% a year until 2046. That coupon puts the software company’s borrowing cost closer to what a heavily indebted corporate borrower pays than to what the largest technology firms have been charged for twenty-year money.
Oracle priced $10 billion of notes on November 2, according to a filing with the Securities and Exchange Commission, with coupons starting at 7.000% and maturities spread from 2029 to 2046. The company said it will use the proceeds for general corporate purposes, including capital spending, repayment of debt and future investments.
The sale had been flagged. At its investor day the previous week, Oracle told analysts it planned to raise $10 billion in common stock and $10 billion in senior notes, part of a financing plan built around the data center construction the company has committed to for its cloud business.
Investors did not treat the pricing as good news. The shares fell 3% in premarket trading on the day the filing landed, according to market data, extending a decline that has taken the stock down 33% this year and cut its value roughly in half over the past twelve months.
The same investor day set a revenue target of $225 billion for fiscal 2030, a figure that assumes the company converts a large book of AI computing contracts into billed revenue over the next four years. The bond sale is one of the bills that arrive before the revenue does.
Long-dated corporate debt is priced against the Treasury curve plus a spread that reflects the issuer’s credit and the market’s appetite for duration. A coupon near 8.65% on a bond maturing in 2046 is unusual for a company of Oracle’s size and rating, and it tells you what bondholders think of the plan: they will lend, but at a price that compensates for the possibility that the buildout takes longer or costs more than the company has said.
The reason Oracle is borrowing rather than paying from cash flow is arithmetic. Building data centers requires land, shells, power contracts, cooling and chips, and the chips are the expensive part, ordered years before a customer’s workload arrives. Cloud providers book revenue when a customer uses the capacity, not when the contract is signed, which means capital goes out first and comes back later, a pattern that strains even a large balance sheet.
Oracle’s position differs from that of the three largest cloud operators in one respect that bond investors care about. Microsoft, Amazon and Google fund most of their construction from operating cash flow, generated by businesses that were profitable before the AI buildout began. Oracle’s cloud infrastructure business is growing quickly from a smaller base, and the contracts it has signed commit it to capacity that has to be financed in advance.
Issuing equity alongside debt is the part of the plan that changes the story. A company that funds a buildout entirely with bonds is betting that the return on the assets exceeds the interest rate. A company that also sells $10 billion of stock is telling shareholders that the debt market alone will not carry the plan at a price it is willing to pay, and asking them to absorb dilution instead.
The 2046 maturity carries the most information for anyone trying to price that trade. Twenty years is longer than any forecast of AI demand deserves to be trusted, and the coupon reflects it. Shorter paper in the same sale priced at 7.000%, a spread over the long bond that pays investors for the extra uncertainty rather than for Oracle’s credit.
Capital spending is now the number that drives the shares. Every quarter the company reports how much it spent on property and equipment, how much of the contracted backlog converted to revenue, and how much depreciation is arriving from assets that were bought for customers who have not yet fully ramped. The market has been sensitive to the gap between the first and the second.
Depreciation matters more than it used to. A GPU cluster has a working life measured in years, not decades, and the accounting assumes a schedule that may prove optimistic if the next generation of chips makes the current one uneconomic to run. Companies that bought heavily at the top of a cycle have historically discovered this in year three, when the assets are still being depreciated and the workloads have moved.
There is a counterargument that deserves weight. Oracle’s contracts with AI developers are long and, by the company’s account, largely non-cancelable, which gives it revenue visibility that most hardware buyers never get. A customer who has committed to buy computing for several years is a form of collateral, and lenders have been willing to treat it that way.
The test will come in the credit market rather than the equity market. If Oracle returns for another large sale and pays a similar coupon, the price of its plan is set. If the spread narrows, bondholders will have decided the buildout is on track. The stock’s reaction this week suggests shareholders are not waiting for that verdict.
For now the company has $10 billion and a schedule of interest payments that runs to 2046. The revenue that justifies both arrives on a timetable the company controls only partly, since it depends on chips it has ordered, power it has contracted and customers whose own businesses are being repriced by the same technology.
