Oracle's Customers Started Paying for the Datacenters

Oracle reported fiscal first-quarter results on September 10, and the headline numbers were the ones the AI trade wants to see. Revenue rose 30% to $19.3 billion. Cloud infrastructure revenue rose 121% to $7.4 billion. Remaining performance obligations, the contracted revenue Oracle hasn't recognized yet, reached $664 billion, up $209 billion from a year earlier and $26 billion from the prior quarter. Adjusted EPS of $1.92 beat estimates.

CEO Clay Magouyrk said Oracle closed more than $30 billion of additional AI contracts in the quarter "without requiring additional capital." CFO Hilary Maxson said the vast majority came through prepayment or bring-your-own-hardware deals.

That sentence is the real news of the quarter. For a year, the bear case on Oracle was simple: it was borrowing to build datacenters for customers who might not be able to pay. Oracle's answer this quarter is a new contract structure. The 10-Q, filed alongside the results, shows what that structure costs.

The capital didn't disappear. It moved.

Oracle spent $28.5 billion on capital expenditures in the quarter ended August 31. A year earlier it spent about $8.5 billion. Operating cash flow was a record $23.1 billion, which leaves free cash flow at roughly negative $5.4 billion for a single quarter.

That operating cash number is flattered by customers. The 10-Q says Oracle received $11.4 billion of customer prepayments during the quarter that "included a significant financing component." Deferred revenue rose from $9.9 billion to $14.7 billion. Put simply, customers are now handing Oracle cash up front to build the capacity they will later rent, and Oracle has to account for part of that as financing.

The rest came from the capital markets. Oracle raised $19.9 billion by selling common stock through its at-the-market program during the quarter, the program it set up in February. Notes payable stood at about $125.3 billion at quarter end. Interest expense was $1.43 billion, up from $923 million a year earlier, an increase of roughly 55%.

So "no additional capital" is accurate for the new contracts, and only for them. The buildout already in the backlog is still being paid for with customer prepayments, a $20 billion stock sale and $125 billion of debt. Full-year capex guidance stays at $90 billion to $95 billion.

Read the backlog's maturity schedule

The $664 billion figure gets quoted as if it were next year's revenue. The 10-Q breaks it down, and the breakdown matters more than the total.

Oracle expects to recognize about 13% of RPO over the next twelve months. That's roughly $86 billion. Another 37% lands in months 13 through 36, 34% in months 37 through 60, and the remaining 16% after that.

In other words, half of Oracle's backlog is not scheduled to become revenue until after August 2029, and about $106 billion of it sits more than five years out. These are take-or-pay commitments from AI labs and hyperscalers whose own economics are still being worked out in real time. A contract signed for 2031 capacity is a promise about the world in 2031.

Near-term guidance is consistent with the schedule. Oracle guided fiscal 2027 revenue to at least $90 billion and second-quarter cloud growth of 64% to 71%. That's rapid growth, but it's a fraction of the backlog. Oracle's future is contracted. What remains uncertain is whether its customers will still be willing and able to pay when the invoices arrive.

Concentration is easing, slowly

OpenAI remains Oracle's defining customer. Oracle doesn't disclose any customer's share of RPO, and estimates in circulation are analysts' numbers, not the company's. Management did say the non-OpenAI portion of the backlog more than doubled over the past year, and Magouyrk said Oracle had expanded the relationship to offer OpenAI's API, ChatGPT for work and Codex through Oracle. Multicloud database revenue, which runs Oracle databases inside rival clouds, grew 353%.

Both are real diversification. Neither changes the core exposure. Oracle is building gigawatts of capacity mostly on behalf of frontier labs, and the timing matters: two days after the call, Sam Altman told Fortune that OpenAI won't go public in 2026. That takes one expected source of fresh capital off the table for OpenAI's side of these contracts, at least for now.

Why the stock went nowhere

The market's verdict was ambivalent. Shares jumped about 7% in premarket trading on September 11, then gave back the gain by the close. According to Yahoo Finance, the stock remained down more than 20% for the year even after a rebound from its July low.

That reaction makes sense. The quarter showed demand is real, capacity is getting delivered (Oracle said it brought on 850 megawatts of AI capacity) and new contracts are being structured to protect Oracle's balance sheet. It also showed the old contracts still need enormous amounts of outside money, and most of the backlog comes due years from now.

What Oracle has actually changed is who takes the risk. Under prepayment and bring-your-own-hardware deals, the customer funds the silicon and Oracle supplies the land, power, construction and operations. That's closer to a utility or a landlord than a traditional cloud. It lowers Oracle's downside per contract, and it probably lowers the margin too. Investors are still deciding whether that's a better business or just a safer one.

What to watch

How much of next quarter's bookings are customer-funded. If prepay and bring-your-own-hardware become the default, Oracle's capex curve should bend before its revenue curve does.

The 12-month slice of RPO. Watch the dollar amount behind the 13% figure, not the $664 billion total. That's the revenue Oracle can actually defend next year.

Interest expense against cloud infrastructure growth. Interest rose about 55% and cloud infrastructure revenue rose 121%. As long as that gap holds, the debt is working. Once it narrows, the bear case returns.