Oracle

Where we land

Oracle's sales rose 22% in a year. The company is now staking what comes next on data centres that rent out computing power for artificial intelligence, with OpenAI, the maker of ChatGPT, as the anchor customer. We are encouraged by how quickly that business is already showing up in sales and profits. The build, however, leans heavily on borrowing: net debt now stands at twice shareholders' equity, meaning the company owes twice what its owners' stake is worth on paper.

We award Oracle a bronze medal.

What the business is

Most people know Oracle as the database company, the software big organisations have used for decades to store and manage their records. That business is still there, but it is no longer where most of the money is. About 77% of revenue now comes from what we class as hosting services, a line that mixes Oracle's newer business of renting out computing power, including the AI data centres, with its long-standing income from supporting software customers and selling business applications delivered over the internet. Our figures do not show how much of that line is AI infrastructure, or how much comes from OpenAI.

The rest is spread thinly: consulting about 9% of sales, software that runs companies' own technology systems another 9%, and server computers the remaining 5%.

The direction of travel is clear. Hosting made up 71% of sales in July 2023, and its sales have grown 55% over roughly the three years since. Consulting has grown faster, by 73%, but from a much smaller base, so its share has only edged up.

This is a profitable business by any ordinary standard. Of every dollar of sales, about 62 cents is left after the direct cost of providing the product, a figure known as gross margin, and about 34 cents remains as operating profit after everything else. Over the past five years gross margin has usually sat around 66%, moving about 4 points either way, so today's level is at the low edge of its normal range, a direct consequence of the business Oracle is choosing to grow.

Why now

For several years Oracle's operating profit went sideways or drifted down, from 15.9bn in the twelve months to August 2021 to 14.3bn two years later. Net profit, what is left after interest and tax, was more erratic still, dropping to 5.8bn in the year to August 2022. Then the line turned. Operating profit reached 18.7bn in the year to August 2025 and 24.6bn in the year just ended, and net profit climbed to 18.9bn. A company that looked mature has started growing like a young one, and the turn coincides with the shift towards renting out computing power.

That is unusual among Oracle's cloud rivals. DigitalOcean, a smaller cloud provider, is growing about as fast, but Oracle's 34% operating margin is the highest of its peers, and investors pay 22 times its yearly earnings, the price-to-earnings ratio, against 61 times for DigitalOcean. Peers on similar multiples, such as Dropbox and IONOS, are growing far more slowly.

What makes the moment pressing is how much rests on a single customer. On 8 October the shares fell after a report that OpenAI's yearly revenue is running near 50 billion dollars, about 20 billion less than earlier reports suggested. Commentators now openly question a story that depends so heavily on one buyer's ability to pay.

The case for

The strongest argument is that growth is speeding up. Total sales grew 21.6% in the last year, against an average of 12.1% a year over the past three. That is rare for a business of Oracle's size and age, and it is coming from the part of the company that now accounts for three-quarters of sales.

Analysts expect the acceleration to carry through to profits. Their consensus sees earnings per share rising about 35% next year, and of recent changes to their forecasts, 24 of 32 were upward. Forecasts are not results, but in the middle of a large investment cycle the direction suggests the people closest to the numbers are seeing contracts turn into revenue.

Those forecasts rest on a business that was profitable before the AI build began, not a start-up hoping to break even. That gives Oracle something many faster-growing peers lack: an established stream of profit to absorb the cost of building.

Where the growth could come from

The shares are trading on one question: can OpenAI pay for the data centres Oracle is building, and can Oracle fund them until it does?

The engine is the AI data-centre business inside the hosting line, and the analysts' expected jump in earnings depends on these contracts turning into paid usage. If OpenAI scales back or stretches out its commitments, that path flattens. Evidence that it is working would be OpenAI using and paying for its contracted capacity on schedule, and other customers signing large AI capacity deals of their own.

The second thread is money. Bloomberg reports that Oracle has nearly doubled its long-term debt over two years, to more than 160 billion dollars. The Wall Street Journal reports talks with Apollo and Goldman Sachs over a deal in which a separate company would buy AI chips and Oracle would lease them, keeping the purchase off its books. Borrowing has also become dearer: the yield on 10-year US government debt, the floor for long-term corporate borrowing costs, has risen to 5.22%, and the Federal Reserve raised its policy rate in September. Higher interest and lease costs eat into the profit on every AI contract, and tight funding could slow the build even if demand holds.

The third thread is physical: getting the data centres switched on. Oracle sent a force majeure notice, a formal warning that events beyond its control may delay its obligations, on its Project Jupiter site in New Mexico because a gas pipeline is late, though it says the project is on schedule. It is reported to be trucking natural gas to sites in Utah and Texas at about four times the cost of pipeline gas. An outside analysis expects its Wisconsin campus to reach full capacity late in 2028 rather than in the second half of 2027, because a high-voltage power line awaits approval. Oracle has also agreed to absorb about 300 million dollars of rising power costs from a Wisconsin nuclear plant, subject to the state regulator's approval.

Two smaller threads could soften the dependence on OpenAI. Oracle is adding AI agents, tools that carry out tasks on a user's behalf, to Fusion and NetSuite, the software large companies use to run finance, staffing and supply chains; a Fusion AI feature was credited with helping the shares recover some losses. Selling AI tools to existing customers needs little new capital, though for now this is an option rather than the engine.

Oracle has also struck deals to run its database inside the data centres of Microsoft, Amazon and Google, so big companies can move their Oracle systems to whichever cloud they prefer without leaving Oracle. Our figures do not show its size, but the growth pattern across the hosting and software lines fits it most likely contributing. Its appeal is that it earns cloud revenue largely on other companies' capital spending rather than Oracle's borrowing. Rapid growth in database usage through rival clouds, with existing customers moving over rather than leaving, would confirm it.

The case against

The scale of the spending is the first problem. Capital spending on buildings and equipment was 28.5 billion dollars in the June-to-August quarter, against 2.3 billion in the same quarter two years earlier, a twelvefold increase paid for largely with debt, and the cash left after that spending is negative. Oracle is not funding growth comfortably from profits; it is borrowing ahead of revenue that has not yet arrived and exploring outside financing to keep going.

The second is who the revenue is meant to come from. If OpenAI is earning about 20 billion dollars less than thought, its capacity to pay for everything it has contracted becomes a real question, and a supplier that has built much of its future around one buyer has little room to manoeuvre.

The third is execution. The New Mexico pipeline, the Wisconsin power line and the trucked gas all point the same way: capacity arriving later and costing more to run than planned. Each month a site waits is a month of interest without matching revenue.

The fourth is quieter. Gross margin has slipped to 61.9%, because renting out AI computing power carries less profit per dollar than software licences and support. Operating profit has still risen strongly, but the mix is shifting towards work that is more expensive to deliver.

Finally, investors have already turned. The share price has fallen 52% over the past year even as profit forecasts have risen, moving the shares noticeably towards the cheaper end of the market relative to earnings. The market is not doubting the forecasts so much as discounting the risk that the build-out fails to pay for itself.

What the price assumes

One way to judge a share price is to ask how fast earnings must grow to justify it. Assume the price stays where it is and that in five years the shares trade at 18.4 times earnings, the middle value for profitable listed companies we cover. Oracle then needs earnings to grow by only about 4% a year.

Its recent record has been far stronger: operating profit has grown about 18% a year over the past three years. If that continued, earnings per share in five years would reach 14.37, against the 7.67 the current price requires, or 187% of what is needed.

This is the past carried forward, not a forecast, and the debate is precisely whether that trend can be sustained on borrowed money. But the price does not demand a triumph, only that the AI build keep profits growing at a fraction of their recent rate. After last year's fall, with the shares up just 2% over six months, the bar has been lowered sharply.

The other side of the medal

The risk that matters most is the combination: a debt-funded build whose payback depends on OpenAI paying on schedule, while power delays and dearer borrowing push costs up. Any one alone would be manageable for a company of Oracle's size. Together they compound, because a late site costs more interest, a weaker customer stretches the payback, and higher rates make every month of waiting more expensive.

The balance sheet leaves little cushion. And although the share price has fallen further than most of the companies we rate, on price relative to assets, sales and earnings the shares remain dearer than most of them, so the market is still paying for a good deal of the growth to arrive.

The risk would show itself as OpenAI delaying, shrinking or renegotiating its commitments, financing closing only on costly terms, and new sites slipping further while borrowing keeps rising. If those arrive together, the forecast jump in earnings does not appear, and the debt remains.

What to watch

The most useful sign would be large AI capacity deals signed by customers other than OpenAI, loosening the dependence on one buyer. Alongside it, watch whether the chip-leasing deal with Apollo and Goldman Sachs closes on terms bond investors accept. Together these would suggest that profit growth of around 18% a year can keep going.

The warning signs run the other way: OpenAI scaling back what it has committed to, or capacity slipping further out while debt and capital spending keep climbing. The New Mexico pipeline, the Wisconsin power-line approval and the regulator's decision on the nuclear power costs are the concrete events that will show which way this is going.

The numbers

The same industry, for comparison

Research for information, not personal investment advice or a recommendation to buy, sell or hold any security: see Important information. Data as at 10 October 2026.