Oracle (NYSE: ORCL) paid for a record quarter of capital spending with new stock rather than new debt. The company sold $20 billion of common stock through an at-the-market program during the quarter, and its total borrowings fell over the same period. Both disclosures sit in Thursday's earnings release, below the numbers that moved the share price.
An at-the-market program feeds new shares into the open market at prevailing prices. It raises cash without a headline offering. It also raises the share count. That is where the quarter's most-quoted figure gets complicated.
Adjusted net income rose 34%, to $5.76 billion. Adjusted earnings per share rose 30%, to $1.92. The gap between those two growth rates is dilution — the effect of spreading the same profit across more shares.
Diluted shares averaged 3.00 billion in the quarter. A year earlier the figure was 2.91 billion. Divide the same adjusted profit by that smaller base and per-share earnings come to $1.98.
Analysts polled by LSEG, the financial-data firm that compiles Wall Street forecasts, had expected $1.74 a share. So the reported beat was 18 cents. On last year's share count it would have been 24 cents. The extra shares did not erase the beat. They took roughly a quarter of it.
One caution belongs with that arithmetic. Weighted-average share counts capture only the portion of a quarter in which a new share existed. The equity sale completed during the quarter, so its full weight has not yet reached the denominator. The larger share base arrives in the next report.
Dilution is not a scandal, and it is not automatically a cost to shareholders. Money raised from new shares can earn more than the slice of ownership it gives away. But the trade only works if the assets bought with that cash earn more than they cost. That test takes years, not quarters.
What the quarter delivered
Revenue grew 30%, to $19.35 billion. The LSEG consensus stood at $19.14 billion. Oracle also cleared its own June guidance, which had capped adjusted earnings at $1.76 a share.
Cloud infrastructure — the rented computing that AI models are trained and run on — did the work. That line grew 121%, to $7.4 billion. Cloud applications, Oracle's older software-subscription business, grew 10%. Software license revenue fell 3%.
The declining software line is not a warning on its own. Oracle attributes it to customers moving workloads off their own servers and onto its cloud. Revenue that used to arrive as a license fee now arrives as rent. The company keeps the customer either way. What changes is the capital required to serve it.
Why It Matters
Oracle spent this quarter proving two things at once. One is that demand for its rented AI computing is real and still accelerating. The other is that meeting that demand costs more cash than the business currently produces. The first claim now has evidence behind it. The second is why the funding question exists at all.
That is what makes the equity sale the quarter's most consequential disclosure. It shows how management chose to close the gap, and who carries the cost of closing it. Lenders carry it under one method. Existing owners carry it under the other. Oracle has switched from the first to the second.
The bill
Capital spending in the quarter was $28.5 billion. Revenue was $19.3 billion. Oracle spent roughly $1.47 on property and equipment for every dollar of revenue it recognized.
That ratio is what separates today's Oracle from the software licensor it used to be. Software companies do not normally outspend their own revenue on physical assets. Data center operators do. The financial statements now read like the latter, and that is a different business with different rules about cash.
Operating cash flow set a record, at $23.1 billion, up 184%. It still was not enough. Free cash flow — cash from operations after capital spending — came to negative $5.4 billion.
Oracle also said it booked more than $30 billion of new AI cloud contracts during the three months. Set that beside the $28.5 billion of capital spending in the same window. The company is laying down close to a dollar of physical capacity for every dollar of new business it signs.
That is the arithmetic a software investor has to absorb. In the license era, the marginal cost of the next contract was near zero. In this era it is a building, a power connection and a rack of chips, paid for before the customer pays. Growth is no longer free, and the balance sheet is where the price shows up.
This is why headline profit growth and per-share growth have begun to diverge at Oracle. One measures the business. The other measures what an existing stake gets. When a company issues stock to build, the two stop moving together.
The funding switch
Total debt fell to $125.3 billion from $129.5 billion at the end of May. That is an unusual direction for a record spending quarter. The $20 billion equity sale is the reason.
It marks a change of method. In FY2026, the buildout was financed almost entirely with borrowed money. Oracle leaned on the bond market and on a convertible preferred issue instead.
That fiscal year produced $55.7 billion of capital spending and negative free cash flow of $23.7 billion. It was funded with $46.1 billion of new senior notes and term loans. This single quarter has already consumed more than half of that annual capital figure.
Cash generation improved on a similar curve. All of FY2026 produced $32.0 billion of operating cash flow. This one quarter produced $23.1 billion — about 72% of the prior year's total.
Part of that jump reflects how AI contracts are paid. Customers prepay for reserved capacity, and prepayment lands in operating cash flow long before the matching revenue is recognized. It is real cash. In accounting terms, though, it funds a liability rather than harvesting a profit. That is how record operating cash and negative free cash flow show up in the same quarter.
Choosing equity over bonds is not a neutral swap. Debt carries a lower cost while rates hold and leaves ownership untouched. But it fixes the repayment schedule regardless of how the buildout performs. Equity costs existing holders a slice of the company and never has to be repaid. A firm expecting a long, uncertain construction cycle has an argument for the second option. A firm confident of near-term cash has an argument for the first.
Neither route is available without limit. Lenders watch coverage ratios. Equity buyers watch how much of the company they are asked to absorb. Oracle has used both channels inside four quarters, which narrows how much of either remains easy to reach.
The demand side
Contracted revenue Oracle has not yet delivered — the company calls it remaining performance obligations — reached $664 billion. That is up $209 billion from a year earlier. StreetAccount's consensus had it at $630.6 billion.
Compared with the prior quarter rather than a year ago, the backlog grew about $26 billion, from $638 billion at the end of May. Oracle recognized $19.3 billion of revenue in the same span. Signings are still outrunning delivery.
The backlog figure invites a fair objection. Contracted is not delivered. A multi-year AI agreement can be restructured, stretched or renegotiated, and backlog has flattered plenty of companies that never converted it. What makes this quarter's version harder to wave away is the operating data Oracle put beside it.
Oracle Cloud Infrastructure President Clay Magouyrk told analysts that GPU utilization ran at 97.9% in the quarter. He said capacity coming up for renewal was renewed or resold at a 20% premium to the prior contracts. The company also reported delivering 850 megawatts of new data center capacity and more than 300,000 GPUs to AI cloud customers.
Utilisation near the ceiling and renewals at higher prices are the two things a landlord wants to see. They are also precisely what would be missing if AI computing were heading into oversupply. That is the case Oracle is making, and this quarter it brought measurements rather than assertions.
There is one thing the utilization figure cannot tell you. High usage of installed capacity says nothing about the price of capacity still to be built, or about what happens when rivals finish their own sites. Tight supply is a condition, not a permanent feature.
What would break this reading
CFO Hilary Maxson gave the full-year picture on the call. "We continue to anticipate $90 billion–$95 billion in CapEx for the full year, with not more than $70 billion in net cash CapEx," she said, according to a transcript published by Investing.com. The gap between those two figures is capacity that customers fund or reimburse. She also described the economics she expects once sites ramp: a free cash flow conversion ratio "of something like 100% to post-tax EBITDA."
That gap is the load-bearing assumption in the entire structure. EBITDA is earnings before interest, taxes, depreciation and amortization. In plain terms, Maxson expects a finished data center to turn nearly all of its operating profit into cash. If the customer-funded portion does not arrive on schedule, the shortfall has to be met from somewhere else.
Here is what would make the funding reading wrong. If the customer-funded share of capital spending arrives as described, the negative free cash flow is a timing effect rather than a structural gap. The equity sale would then look opportunistic rather than necessary. Magouyrk pointed that way himself. He said Oracle closed more than $30 billion of AI contracts in the quarter "without requiring additional capital from Oracle."
Guidance moved up. Oracle now expects at least $90 billion of revenue and $8.10 of adjusted earnings per share for fiscal 2027. Both figures sit modestly above where the LSEG consensus already stood.
The stock closed the regular session at $152.94, down 5.4%, according to Investing.com.
