A quarter ago Oracle’s cloud infrastructure revenue grew 93%, and this page called the demand question largely settled. It has now grown 121%. What changed is not the demand. One quarter of building cost more than half of last year’s entire build, and Oracle became the first of the five big AI builders named by the Bank for International Settlements (the BIS) to sell new shares to pay for it.
| Line | Expected | Delivered |
|---|---|---|
| Q1 total revenue | ≈$19.14B (analysts) | $19.3B — up 30% |
| Q1 adjusted earnings per share | $1.74 (analysts) | $1.92 — up 30% |
| Q1 earnings per share, standard accounting | — | $1.56 — up 55% |
| Total cloud revenue | +58–64% (Oracle’s own guidance) | $11.6B — up 62%, inside its guidance |
| Cloud infrastructure (renting out computing power) | — | $7.4B — up 121%, from up 93% last quarter |
| Cloud applications (software rented online) | — | $4.2B — up 10% |
| Software | — | $5.6B — down 3% |
| Orders signed but not yet delivered (RPO) | — | $664B — up $209B on a year earlier, up $26B from $638B in the quarter |
| Q1 capital spending | — | $28.5B — from $8.5B a year earlier |
| Q1 operating cash flow | — | $23.1B — up 184%, a Q1 record |
| Q1 free cash flow | — | −$5.4B — from −$0.4B a year earlier |
| Capital spending after customer prepayments | — | $18.0B — Oracle’s own measure, after subtracting $11.4B paid in advance by customers |
| New shares sold in the quarter | ≈$20B planned for the year | $20B sold on the open market, completed |
| Revenue guidance, year to May 2027 | $90B | raised to at least $90B; adjusted earnings per share $8.10 |
| Capital spending plan, year to May 2027 | $90–95B gross | unchanged; about $70B after customer prepayments |
| Capacity delivered since the May quarter | — | 850 megawatts of data centre; over 300,000 AI chips (GPUs), nearly triple the previous quarter |
Analysts’ forecasts are as recorded before the results, and nothing on this page depends on them. Every revenue, earnings, order-book and cash-flow figure is Oracle’s own, from the results release it filed with the US securities regulator on 10 September. The share move is reported in the press rather than filed, and is stated as a range. The three clocks above are the site’s standing questions: is the AI revenue growing fast enough, is the company still committed to the spending, and are investors still willing to wait.
The one thing this page expected to be able to say was that a growth rate this high must start to slow. Growth rates at this scale always do. The revenue-floor argument on the front page (the revenue the AI spending must earn every year to pay for itself) rests on about 40% a year sustained for five years, which is hard precisely because high early growth rates fade. Oracle’s did the opposite: 93% became 121%. On the demand question alone, this is the strongest result any of the five has produced.
The funding question moved the other way, and faster. Capital spending, the money spent on chips, buildings and power, was $28.5 billion in this one quarter. In the whole of the previous financial year, a year that already shocked the market in June, it was $55.7 billion. Oracle is now spending at a pace where three months cost more than half of the previous twelve. Free cash flow, the cash left after capital spending, was minus $5.4 billion, against minus $0.4 billion in the same quarter a year earlier.
What is new is how the gap was filled. In June, Oracle paid for the build-out with debt: $43 billion of it during the last financial year. This quarter it completed a $20 billion sale of its own shares. No other member of the five has done that. Debt is a claim on future cash. Shares are a permanent stake in the company, sold at whatever price the market will pay, here to buy hardware that lasts a few years. A company does that when the build-out cannot wait for either the business or the bond market to pay for it.
Then there is the record operating cash flow, the cash the business brought in. It is the most interesting number in the release and the easiest to misread. Oracle brought in $23.1 billion, up 184%. But $11.4 billion of that, nearly half, is customers paying in advance for capacity that does not exist yet. Oracle says so itself, and subtracts it to reach a “net cash outlay for capital expenditures” of $18.0 billion. On the strain chart on the tracking page, which divides each company’s capital spending by its operating cash flow, this quarter pulls Oracle’s reading down from 174% to 161%. That looks like relief and is not. Take the prepayments out of the operating cash flow and the same twelve months read about 244%. The chart plots the figures as filed and now says so beneath it.
That prepayment is also the clearest instance anywhere on this site of the loop, in which the firms of the AI boom finance each other. The demand and the funding are the same dollar, moving once: a customer commits to buy AI capacity, hands over the cash to build it, and Oracle books that cash as operating cash flow. From the outside, signed demand and self-financing look identical, and the cash-flow statement cannot tell them apart.
First and largest: $664 billion of orders signed but not yet delivered is a promise, not a payment. It is now nearly ten times Oracle’s annual revenue. The release does not break it down by customer, so the concentration that press reports have repeatedly attributed to a small number of AI labs cannot be measured from the filing, only noted. An order book concentrated in customers who have no revenue of their own yet is a different asset from one spread across an economy, and nothing in these results says which this is.
Second, the 300,000 chips are the answer to what delivering that order book costs, and they are only the first instalment. Oracle delivered 850 megawatts of capacity in three months. The spending plan of $90–95 billion for the year, held unchanged, is at or above the $90 billion of revenue guidance for the same year. Oracle is the only one of the five whose capital budget and revenue are the same size.
Third, most of Oracle’s revenue is still not AI revenue. Software fell 3% to $5.6 billion and cloud applications grew 10%. The $7.4 billion infrastructure line growing at 121% is the part the bet rests on, and it is still well under half of a $19.3 billion quarter. Growth of 121% in the smaller part of the business is consistent with the bet working and with it failing. What it rules out is demand having stalled.
The quarter to November 2026, expected in the second week of December. Oracle’s own guidance is total revenue growth of 30–34% and cloud growth of 65–71%, so the demand question gets asked a fourth time. Three figures matter more than the growth rate. Whether quarterly capital spending stays near $28 billion, which would carry the year well past the $90–95 billion plan. Whether the customer prepayments repeat, since the operating cash flow line now depends on them. And whether Oracle sells more shares, having said it did not expect to issue more debt in calendar 2026. Oracle is the one member of the five whose results the site’s data pipeline does not capture, so its figures are checked directly against its filings.