Reference price: $163.85 — September 11, 2026 at midday, after first-quarter results; the stock had closed at $152.94 on the 10th. Market value around $491.55B, counting the 3,000 million shares outstanding after this summer's share sales. Fiscal years end May 31. Prices move in real time.
The verdict
Not a buy. Oracle sells more every quarter and has more signed orders than almost anyone on Earth. But to deliver them it spends far more money than it takes in — and that money is being put up by everyone who walks past: bondholders, new shareholders, and even its own customers.
What this company is
What it was: the database company — the program where banks, airlines and hospitals keep everything that matters. A quiet, very profitable business.
What it is now: a landlord of supercomputers with a software company inside. It builds halls full of processors for artificial intelligence and rents them out to a handful of enormous clients.
The stock listing still says "software". That's misleading. The business that's growing looks more like a utility than a program: you spend the money on the power plant first, and collect it back over years.
The good
Sales up 30%, and the new business more than doubled
| Q1 of the fiscal year (June–August 2026) | Sales | vs. a year ago |
|---|---|---|
| Total | $19,345M | +30% |
| Compute capacity rental | ~$7,400M | +121% |
| Cloud applications | ~$4,200M | +10% |
| Legacy software | $5,550M | −3% |
Why it matters: a company this size doesn't usually accelerate. This one went from growing 17% in the previous fiscal year to 30% this quarter.
$664 billion in signed orders — more than eight and a half years of today's sales
In a single quarter it signed more than $30 billion in new contracts, and the backlog has grown $209 billion in a year. The company says demand is growing faster than its capacity to serve it.
But that backlog doesn't arrive at once. Of the $638B signed as of May 31, its own annual report says it will be billed like this: 12% in the next 12 months, 34% between years 1 and 3, 34% between years 3 and 5, and 20% after that.
The forecast is unusually well backed
Analysts expect $8.08 of earnings per share this fiscal year. The company guides $8.10. They agree — and the estimate is signed by many: 37 analysts for 2027, 37 for 2028, 21 for 2029, 11 for 2030. Every year clears ten. It's one of the best-supported forecasts you can find.
And this quarter, profit kept up: business profit before one-offs rose 31%, about the same as sales. It ties; it doesn't win. What's hiding inside that tie is Problem 5.
The bad
Problem 1: it spends nearly twice what comes in
| FY ended May 2026 | Q1 of the new FY | |
|---|---|---|
| Cash in from operating | $31,977M | $23,103M |
| Cash out on building | $55,663M | $28,499M |
| What's left | −$23,686M | −$5,396M |
In the twelve months to August, after also paying its people in stock, it is short about $33.5 billion. And it still pays a dividend: $5,787M last fiscal year, now $0.50 per share per quarter — paid with money it doesn't have. Concretely: a broker loan costs 5–6% a year. A share with a negative yield doesn't pay that rent — it charges it.
Problem 2: half of this quarter's cash was lent by its customers
Of the $23,103M that came in from operating, $11,363M is customers paying in advance, with interest — the company itself labels them payments with a "significant financing component". Another $3,997M is ordinary subscription prepayments.
Without the borrowed $11,363M, operating cash falls to $11,740M: half. And in its own tables, Oracle uses those advances to pay for construction. That money isn't profit. It's a debt to the customer, repaid in service. While new contracts keep getting signed, it looks like cash. The day they stop, that tap closes.
Problem 3: it is raising money through every door at once
| In about fifteen months | Amount |
|---|---|
| New bonds | $43,000M |
| Preferred shares that will convert to common (at 6.5%) | $4,954M |
| New shares sold on the market, in a single quarter | $20,000M |
| Customer advances carrying interest | $15,955M |
| Total | ~$84,000M |
The share-sale program it opened in February was $20B. It exhausted it in one quarter. And the release says its plans to keep raising capital are unchanged. Every new share splits the profit between more people: a year ago there were 2,909 million shares; now there are 3,000.
Problem 4: $125B of debt — plus $260B signed that isn't on the balance sheet yet
| As of August 31, 2026 | |
|---|---|
| Bonds and loans | $125,337M |
| Cash and securities | $37,077M |
| Net debt | $88,260M |
| Interest for the quarter | $1,428M (+55%) |
That debt is serviceable: business profit covers interest 4.5 times. The problem is what's not in that table. The annual report recognizes $260B in already-signed data-center leases that don't appear on the balance sheet yet. They start between this year and 2029 and run fifteen to nineteen years. Spread out, that's on the order of $14–17B of rent per year once they're all running. Add $19B more in equipment purchases signed after May, and a guarantee of up to $3.3B on a landlord's debt, expiring September 2026.
Problem 5: every new dollar leaves less behind than the old ones
| Cloud & software business | A year ago | Now |
|---|---|---|
| Of each $100 sold, left after direct costs | $72 | $63 |
| Equipment wear per quarter | $1,351M | $3,156M (+134%) |
| Equipment and buildings on the balance sheet | $53,194M | $127,845M |
Why it matters: those $127,845M of halls and processors will be subtracted from profit, year after year, for as long as they last. Renting compute leaves far less than selling software. Long term it already shows: in 2017, 34 of every $100 of sales survived as business profit; in 2026, 31.
Problem 6: the screener's profit is 31% higher than the official one
For the fiscal year ended May 2026: the panel says $7.63 per share; official accounting says $5.83. The difference is mostly the ~$4.8B a year it pays employees in stock, which the panel number doesn't count as an expense. It is a real expense: the shareholder pays it, watching the profit get split between ever more shares.
Problem 7: to deliver the 2030 promise, sales must nearly double in three years
Analysts promise $19.83 per share for fiscal 2030. Taken to official accounting with today's share count, earning that requires billing about $163B a year — at the best margin it has ever achieved, from 2021. This year it expects to bill at least 90. That means growing 22% a year for three straight years. Hard, but not impossible — unlike almost every case, here much of it is signed: split the backlog evenly and that year would get about $108B from contracts that already exist. About $55B is missing — from contracts that don't yet exist, and from the legacy business.
And the backlog depends on few clients. Its annual report says so: part of the cloud business is "more concentrated in a small number of large customers". If one of them can't pay, that backlog shrinks.
The sum of it all — only one of two things can happen
The usual math doesn't work here, and it's worth saying why: the standard screen fails it (about 21 times earnings, growing 5.9% this year — it fails year one); the margin of safety can't be computed (to value a business you discount the money it has left over — Oracle doesn't have money left over: it's missing money); and the reverse calculation is brutal: with debt included, the company is priced around $580B. To justify that without ever growing again, it would need to leave about $52B a year free, forever. Its best year in history was $11.9B. Four and a half times more, sustained.
So only one of two things can happen:
- Either it keeps building — and the money keeps flowing out negative, with more debt and more new shares;
- or it stops building — and the growth that justifies the backlog stops with it.
Both at once, no. And today's price is paying for both.
Price against profit
The middle column is profit multiplied by fifteen — the "normal" price of a healthy company. The distance between that column and the price is how expensive or cheap it is.
| Fiscal year (ends May) | EPS (panel) | What it would justify | Price at FY close |
|---|---|---|---|
| 2019 | $3.52 | $52.80 | $50.62 |
| 2020 | $3.85 | $57.75 | $53.78 |
| 2021 | $4.67 | $70.05 | $78.74 |
| 2022 | $4.90 | $73.50 | $71.93 |
| 2023 | $5.12 | $76.80 | $105.94 |
| 2024 | $5.56 | $83.40 | $117.19 |
| 2025 | $6.03 | $90.45 | $165.53 |
| 2026 | $7.63 | $114.45 | $225.78 |
| Today | (2027: $8.08) | $121.20 | $163.85 |
| 2028 promised | $11.02 | $165.30 | |
| 2029 promised | $15.35 | $230.25 | |
| 2030 promised | $19.83 | $297.45 |
From 2019 to 2022 the price is deduced from the starting panel's multiple; from 2023 to 2026 it's the real close of each fiscal year, matching the panel to the cent.
For years the price hugged that column — 14 to 17 times earnings. In 2025 and 2026 it detached: it reached 30 times, peaking at $328.33 on September 10, 2025. Since then it has fallen by half — down to where the 2028 promise sits ($165.30). Today you pay the profit of two years from now. And under official accounting, profit is 25–30% lower than the panel's. This whole table is built with the good number.
What we would do
Not buy. And in this case the signal isn't a price — it's a sign: the day the money coming in from operating exceeds what it spends on building without counting what its customers lend it.
What you would be paying
| Price | What it is |
|---|---|
| $225.78 | Where the fiscal year closed, May 29, 2026 |
| $165.30 | Fifteen times the profit promised for fiscal 2028 — practically today's price |
| $163.85 ← it is here | September 11, 2026, after results. The day before it closed at $152.94 |
| $121.20 | Fifteen times what's promised for the current year |
| $114.99 | Its twelve-month low, July 24, 2026 |
The average analyst price target is $244.37. That's their number — and it's built with the panel's profit.
The three things we would watch
- The sign, without customer advances. Operating cash, minus construction, minus what customers lend it. In the next results (around December 10, 2026) it should stop being negative. If it doesn't, Problem 1 is alive.
- The next share sale. It exhausted $20B in one quarter and says its capital-raising plans are unchanged. Every new round splits the profit between more shares.
- The $63 out of every $100. If what's left from each cloud sale keeps falling while the halls fill up, the 2030 promise moves further away. And the $3.3B guarantee was due to expire in September 2026.
In one sentence
Where every figure comes from
First-quarter results published September 10, 2026; annual report for the fiscal year ended May 31, 2026; historical accounts in the SEC registry; daily Nasdaq prices; Finviz estimates and quotes plus the starting panel. The price moves in real time: $163.85 is the September 11, 2026 midday quote.