Oracle's Growth Is Contracted and Real. The Cash Flow, Less So — What Thursday's Report Signals

Oracle reports its fiscal first-quarter results Thursday after the close, and for once the calendar matters. For the past year the company has been telling one story and the stock has been telling another, and this report is where the two have to meet.

Here is the gap. On the operating side, Oracle ended last fiscal year with $638 billion of signed but unearned revenue in its backlog, up more than threefold from a year earlier—a pipeline large enough that management guided the current year to roughly $90 billion of revenue, about 34% above the year before. Cloud infrastructure, the engine of the buildout, grew 93% in the final quarter. On the tape, meanwhile, the stock sits near $159, down roughly a third over the past twelve months, about half its 52-week high near $346, and down 18.5% year to date. A business that keeps beating expectations and raising guidance has been marked down all year. That divergence—improving operating numbers, retreating price—is the whole question in front of Thursday's print, and it is not a crowded one.

A rocket that burns cash while it flies

The reason the market is not paying for Oracle's growth has little to do with demand. It has to do with what the growth costs. The AI data-center buildout that produces those revenue numbers is the most expensive capital project in Oracle's history. Last fiscal year the company spent $55.7 billion on capital expenditures—more than 2.6 times the $21.2 billion of the prior year—and guided fiscal 2027 to a gross buildout of $90 billion to $95 billion, or net project outlay of roughly $70 billion after customer prepayments.

The result is that Oracle, for all its revenue growth, does not generate free cash flow at all. Free cash flow for fiscal 2026 was negative $23.7 billion: about $32 billion of operating cash flow went out the door against the giant infrastructure check. On top of that it is leaning on its balance sheet, ending the year with about $219 billion of total debt after raising $43 billion in debt and $5 billion in equity during the year, with more financing planned.

This is the mechanism, and it is important to be precise about it. A company can post record revenue and still consume cash every quarter, because cash is spent building the data centers before the contracted revenue lands. That is exactly Oracle's situation. It is also why the June report mattered the way it did: Oracle beat revenue and earnings estimates for the fiscal fourth quarter yet the stock fell more than 9% in a day to $182—investors were focused on the capex overshoot, not the beat.

The backlog is the ballast—and the bet

So the case for Oracle cannot run through free cash flow, and I want to be honest about what that means rather than pretend otherwise. Free cash flow is my preferred hard proof of a story, and here it is the weak point, not the strength. When a thesis has to be carried without it, the uncertainty is genuinely higher, and this one has to be.

The anchor that carries it instead is the backlog, and it is the reason the bear story is incomplete. Remaining performance obligations—contracted revenue that has been signed but not yet recognized—reached $638 billion at the end of fiscal 2026. Roughly $300 billion of that traces to Oracle's landmark five-year deal with OpenAI, struck through the Stargate venture with SoftBank, a contract valued at about $60 billion per year once it begins in 2027. Some $75 billion of the backlog is prepaid or customer-supplied hardware, money customers have put up in advance that reduces how much Oracle itself has to fund. That backlog is the difference between a hand-wave about "AI demand" and a specific, signed set of commitments Oracle can point to for years ahead.

Seen that way, the setup is the expectations-reset profile I look for: the market priced the old, more modest Oracle through last year, celebrated the inflection briefly, and has since swung to treating the capex as irresponsible. The operating numbers, measured in contracted revenue and a cloud line compounding at 50%-plus, have not broken. But unlike my usual read, the proof is not free cash flow showing up soon—it is a backlog converting to revenue and cash over several years. That is a real distinction, and it should lower the comfort level, not raise it.

The strongest bear argument, stated plainly

The bull case rests on a single, large counterparty, and that is the fair thing to press. The OpenAI contract alone is roughly half the backlog and about $60 billion a year once running. OpenAI's own annual revenue reached only about $25 billion, and its chief executive has said the company does not expect to be profitable until around 2029. The gap between the compute Oracle is building and OpenAI's current ability to pay for it is enormous; it is only bridged by OpenAI continuing to raise enormous amounts of capital, as it did with a $122 billion round earlier this year. If OpenAI's funding tightens, or it builds out its own capacity and needs less of Oracle's, a big slice of that $300 billion is at risk. That is the clean break condition, and it is a genuine one, not a scare.

There is a second reason to distrust the apparent cheapness. At roughly $159 and fiscal 2027 adjusted earnings guided to about $8.05 a share, Oracle looks like a technology value stock at around 20 times forward adjusted earnings with 34% revenue growth. But that adjusted figure excludes a great deal of the depreciation of the very data centers creating the risk, and on a GAAP basis the forward multiple is closer to 37 times. The "20 times" is therefore not as cheap as it appears; it rests on earnings that leave out the cost of the assets doing the work. That matters because the whole rerating depends on those assets eventually generating cash, not just booked earnings.

What Thursday's report decides

The report on Thursday, then, is a checkpoint on a multiyear proof path rather than the proof itself. The street is modeling about $19 billion of revenue—up from $14.9 billion a year ago, when the stock shot up a record 36% in a single day after that first AI-infused report—and adjusted earnings of roughly $1.30 a share, a bit under last year's $1.47. That earnings dip is the footprint of the depreciation and interest stacking up from the buildout, and it is worth reading as such rather than as a breakdown. Revenue roughly doubling against a flattish-or-lower earnings number is exactly the shape this story has to show.

The things that matter most are forward-looking. Does management keep the $90 billion revenue target and the capex plan where they are, or does the gross buildout climb again? Does backlog keep growing, and are the customer prepayments holding up—those reduce the money Oracle must fund itself? And what is said about the OpenAI and Stargate relationship, given it is the single largest line in the backlog?

For someone deciding whether Oracle is a real opportunity or a value trap, that is the frame: this is not a free-cash-flow story you can close your eyes and trust. It is a contracted-revenue story whose entire outcome turns on two things that can be watched—whether the backlog converts to cash as revenue scales, and whether the giant spending peaks as a share of revenue rather than climbing forever. If capex overshoots again even as revenue is guided up, the market is right to be skeptical and the stock stays punished. If the burn looks set to peak into that contracted revenue, the market is still pricing the old risk profile while the operating setup is getting cleaner. Thursday gives you one more data point on which way that balance is tipping, and the honest answer is that nobody knows the direction until the conversion actually shows up in the cash flow.