Expand menu

On September 10, after the market closed, Oracle released its first-quarter fiscal 2027 results. The number that made the headlines was the backlog: $664 billion in remaining performance obligations, $209 billion more than a year ago and $26 billion more than in June. That is signed contract that has not been delivered yet.

The same release carries another number. Free cash flow for the quarter was negative: minus $5.4 billion. Twelve months earlier it was also negative, but only by $362 million.

Both numbers are correct, and they describe the same business. It is worth understanding why, because the mechanism is not unique to Oracle.

What happened

Total revenue of $19.3 billion, up 30% year over year. Cloud grew 62% to $11.6 billion. Within that, infrastructure — the business of renting compute to train and run models — grew 121% to $7.4 billion in a single quarter. Cloud applications grew 10%. Traditional software fell 3%.

GAAP earnings per share came in at $1.56, up 55%. Operating cash flow hit a record $23 billion, up 184%.

And capital expenditure for the quarter was $28.5 billion. That is the number that turns a record operating cash flow into negative free cash flow.

For the full fiscal year, Oracle expects to spend $90 to $95 billion on capex against revenue of at least $90 billion. In other words: it plans to spend roughly as much building as it expects to bill. To fund it, the company plans to raise about $40 billion in debt and equity over the year.

Reuters reports that roughly half of the backlog, about $332 billion, is expected to convert into sales within 36 months.

The market did not punish it: the stock rose 7% after Thursday's close and 3% at Friday's open. But it is still down 21% year to date.

Why it matters

A backlog is a promise to collect. It is not a collection.

In classic software, that distinction was almost an accounting footnote. You signed a license, the marginal cost of serving it was minimal, and the gap between signing and collecting was covered by ordinary working capital.

In AI infrastructure, the distinction is structural. To serve the contract you have to buy the GPUs, power the data center and connect it, before you bill the first month. Oracle delivered more than 300,000 GPUs in the quarter. That spending happens today. The collection is spread across the years of the contract.

Which means the faster the backlog grows, the more negative cash flow becomes. Commercial success and cash pressure move in the same direction, not opposite ones. It is counterintuitive, and it is what makes this cycle look unlike any previous software boom.

Morningstar's Luke Yang put it this way: "Despite Oracle asking customers to partially fund technical hardware to alleviate cash flow pressure, we do not foresee Oracle's cash flow profile changing anytime soon."

What it means if you run a technology company

Oracle can carry that gap because capital markets will lend it $40 billion. That is the part of the story that does not transfer.

If you run a services or product company signing large contracts with heavy delivery ahead of them, the mechanism is identical and the safety net is not there. I have seen projects where the team celebrates a big annual contract and six months later the company is opening a credit line to make payroll for the very people delivering it.

Three things change depending on how the deal is structured.

First, the shape of the collection curve matters more than the amount. A $1.2 million contract collected in twelve monthly payments is a different business from the same contract collected 90 days after final delivery. The second one can break a company that the first one would have grown.

Second, the upfront payment is not a favor you ask for — it is part of the design. Oracle is asking its own customers to partially fund the hardware. If Oracle does it, there is no reason a twenty-person company should treat asking for 40% upfront as a commercial weakness.

Third, you need to know which number actually constrains you. For Oracle it is access to capital, not demand. For most other companies it is the same, except they find out late: when the pipeline looks better than ever and the bank account looks worse than ever.

My read

What strikes me most about this report is not the size of the backlog. It is that Oracle published it alongside negative free cash flow without trying to bury it, and the market absorbed it.

That tells me the industry has implicitly accepted that AI infrastructure gets financed the way a dam or a port gets financed: heavy capital first, long collection after. That is a change in kind. Software was a margin business; compute capacity is a balance-sheet business.

For those of us building software and AI systems for companies, the practical lesson is smaller and more urgent: when a client asks for a large implementation with infrastructure in the middle, the conversation about the payment schedule is as technical as the conversation about the architecture. And it has to happen at the same time, not after signing.

A large backlog is not a healthy company. It is a company with work ahead of it, and an open bet on whether it can finance that work.