Oracle Corporation ORCL

Key data

The business

Oracle sells the database software that runs the world's ledgers, the record system embedded in banks, airlines, telecoms and governments for decades, and it now rents the computers those and newer AI customers run on. Cloud and software is 87% of the $67.4 billion in sales, hardware 5%, services 8%. The line that matters is the mix inside it: cloud revenue reached 51% of the total in fiscal 2026, up from 43% and 37% the two years before, as customers on old software-support contracts move onto Oracle's cloud. The moat is the database itself. Ripping it out means re-plumbing everything built on top, which is why support renewals hold. The new engine bolted onto that base is OCI, renting raw computing capacity to AI companies that need it faster than they can build it.

The numbers

Five quarters show a business accelerating and a balance sheet taking on load.

Revenue climbed every quarter to a seasonally large $19.2B in May, up 20.6% on the year-ago quarter. Distrust the November quarter's $6.1B net income and $2.10 EPS: operating income that quarter was $4.7B, roughly in line, and the profit spike came from an investment mark, not operations. On the non-GAAP earnings the Street watches, Oracle beat estimates in three of the last four quarters.

Over five years revenue compounded about 12% a year and earnings per share nearly 25%, operating income roughly doubling to $20.6B. This is a genuinely faster Oracle than the flat company of a decade ago. But the third table is the whole argument.

Operating cash flow rose to $32B, but capital spending hit $55.7B, so free cash flow was negative $23.7B, the second straight year without a surplus after a decade of $5B to $12B. Net debt is about $125B, 3.9 times EBITDA. To fund the build, buybacks fell from $16.2B in FY2022 to $95M in FY2026, essentially off. The market already knows the backlog is large. What it is repricing, after the 54% drop from the high, is whether that capex converts to cash before the debt starts to matter.

Management

No insider has bought a share in twelve months. Insiders sold about $125 million across 27 sales, led by Jeffrey Henley's roughly $24 million in June 2026 and Clayton M. Magouyrk's $11 million last October. Plan status is not disclosed, so read those as neither alarm nor comfort. The louder signal is capital allocation: Oracle, historically one of the market's most aggressive repurchasers, stopped almost entirely and redirected every dollar into data centers. That is a coherent move if the capacity fills and a costly one if it does not. Founder Larry Ellison remains the company's largest shareholder and driving force, so the buildout reflects an owner's conviction, not a hired manager's caution.

How it fails or surprises you

The build outruns the cash. Oracle spent $55.7B on capacity in FY2026 against $32B of operating cash, a $24B gap funded by debt at 3.9 times EBITDA. FY2027 capex against operating cash flow has not printed yet, and the FY2026 close already shows the gap wide open. If capex keeps climbing while cash lags, the balance sheet, not the business, sets the ceiling.

A few customers carry the backlog. The signed contracts that justify the spend lean heavily on a small number of AI companies whose own funding is unproven. If one delays, renegotiates or fails, the backlog the whole story rests on is worth less than its headline. The first tell is a quarter where cloud revenue undershoots the backlog's implied pace. This is the fact the read explains least: the backlog is only as good as its counterparties.

The capacity fills and cash floods back (right tail). If the centers come online and the backlog converts on schedule, cloud revenue reaccelerates past 17% and the capex wave crests, flipping $24B of negative free cash into surplus within two to three years. The market, pricing funding risk after a 54% drawdown, is not paying for that today. The print: cloud growth turning up toward 30% while capex growth decelerates in FY2027.

Closing thoughts

This is the middle world of the three: a specific print settles it. Oracle is not a question of whether the business works, it plainly does, nor a pure survival bet, since the support base services the debt. It turns on one conversion: does the $56 billion of building become recurring cash before the $125 billion of net debt and its interest become the story. The tell it is working is a quarter where capex growth slows and cloud growth holds; the tell it is not is capex climbing again as cloud growth fades, and the reader waits for the next quarter rather than guessing. The downside tail is fatter today, because the debt is real and dated while the payoff is only promised, and what is at risk is not bankruptcy but years of negative free cash and a stock that already remembers $345.

The bet is still that the giant contracts Oracle signed to rent out AI computing power turn into real cash, and keep coming. What breaks it is the funding gap outrunning that conversion. The one pair of numbers that tells you first is FY2027 capital spending set against operating cash flow, and whether the backlog holds its line. If capex crosses another year above cash with the backlog flat, the read is wrong.