Here is the contradiction most Oracle shareholders are living inside without noticing it.
The company's finance and product leadership spent the quarter telling you that new contracts arrive without new cost. Clay Magouyrk put it plainly: "We closed more than $30 billion of additional AI contracts in Q1 without requiring additional capital from Oracle."
That is meant to be a comfort. It is actually the tell.
You cannot both be spending more than you earn and be adding contracts that cost nothing. One of those two statements is hiding the other.
Because the same cash statement that quote was designed to narrate past shows capital expenditure of $28.5 billion in a single quarter — up 235% year over year — a figure that outran Oracle's entire operating cash flow for the period. Operating cash flow hit a record $23.1 billion. It wasn't enough. Free cash flow landed at negative $5.4 billion, against roughly negative $0.4 billion a year earlier.
You cannot both be spending more than you earn and be adding contracts that cost nothing. One of those two statements is doing the work of hiding the other.
Last week the open question in this letter was abstract: does ARR and backlog convert into durable, margin-accretive cash flow, or does it just pile up as a promise? This week we have a real print to test it against — and, for the first time, a named counterparty carrying half the risk.
Start with the headline everyone cheered. Revenue rose 30% to $19.35 billion. Cloud infrastructure revenue more than doubled — up 121% to $7.4 billion. Adjusted EPS came in at $1.92 against $1.74 expected. The stock popped as much as 7% after hours.
It was, by the framing one analyst used, an 85-point quarter against a 70-point hurdle. Every visible line beat. The demand is not in doubt.
Now reframe the number that everyone calls the bull case. Remaining performance obligations — RPO, the contracted revenue Oracle has signed but not yet delivered — stood at $664 billion, up $209 billion in a year.
Investors have been trained to read that as an asset. It is more accurately an obligation.
RPO is not cash in the bank. It is a set of promises Oracle must fund up front — build the data centers, secure the power, install the GPUs, operate all of it — before a dollar of that backlog converts to revenue. The revenue comes years later. The spend comes now. That is precisely why free cash flow is negative while bookings are exploding.
And here is the detail that turns a financing story into a solvency story: roughly half of that $664 billion RPO is tied to a single customer, OpenAI.
OpenAI reported a $20.9 billion operating loss in 2025.
Sit with the structure of that. The largest single counterparty to Oracle's backlog — the buyer underwriting somewhere near $330 billion of contracted future spend — lost nearly $21 billion last year. And it is not pulling back. OpenAI has reportedly raised its 2030 compute spending forecast to $750 billion, up more than 25% from a prior estimate near $600 billion, while still describing its own capacity as "seriously insufficient."
So the bet shareholders are being asked to make is not really a cash-conversion bet. It is a bet that one deeply unprofitable customer keeps writing enormous checks on schedule through the end of the decade.
Management's counter to all this is the prepayment story, and it is a real point worth stating fairly. Most of the recent RPO growth came from large AI contracts where the customer prepaid for the GPUs or supplied the hardware directly. The prepaid and customer-supplied hardware portion now totals $75 billion. Oracle's own language: this "substantially reduces the amount of capital Oracle must raise."
Reduces. Not eliminates. And here is the arithmetic that management's phrasing walks you around: that $75 billion is a cumulative figure, and it does not cover a single year of the build.
Oracle guided to roughly $70 billion in net capital expenditure for fiscal 2027 — plus another $20 billion to $25 billion in component prepayments on top. So the entire cushion of customer-financed hardware, accumulated across all these deals, is smaller than one year of planned spend.
Everything above that line is financed by Oracle. Last fiscal year that meant raising $43 billion in debt and $5 billion in equity. This year the plan is roughly $40 billion more, including a $20 billion at-the-market equity program the company already completed this quarter — which is another way of saying shareholders paid for part of this build through dilution.
Total debt now sits around $125 billion, and Oracle is reportedly the largest non-financial issuer in the U.S. high-grade corporate bond index. Every incremental billion of that carries a cost, with the 10-year Treasury reportedly near 4.7%, characterized as the reference rate on the financing that keeps the buildout moving.
The full-year picture is starker than the quarter. Fiscal 2026 closed with negative $23.7 billion of free cash flow on $55.7 billion of capex. On a trailing-twelve-month basis, Oracle reportedly generated roughly $47 billion in operating cash and spent about $76 billion building — a $29 billion gap that had to come from somewhere.
And it gets worse before it gets better. Some sell-side models don't see free cash flow turning positive until fiscal 2029.
That is the real time horizon of this trade. You are not underwriting a quarter. You are underwriting three-plus years of negative free cash flow, funded by debt and equity, backed by a backlog whose largest single line is a customer losing $20 billion a year.
To be clear about what is not the risk: it isn't demand. GPU utilization ran at 97.9%. Oracle delivered more than 300,000 GPUs and added 850 megawatts of capacity in the quarter. Management reportedly guided total cloud revenue to grow 65% to 71% next quarter, per ainvest, and raised full-year revenue guidance to at least $90 billion. The pipeline is booking as promised.
The thesis does not break on a growth slowdown. It breaks on a conversion slowdown — or, more precisely, on concentration. It reportedly breaks the day one large buyer flinches.
Because when half your backlog rides on one counterparty, the question stops being "will AI demand hold across the market" and becomes "will this specific customer keep paying." Diversified backlog is resilient. Concentrated backlog transmits a single customer's problems straight onto your balance sheet — and Oracle is funding the build against that backlog in advance, with borrowed money, at a reported 4.7% carry.
The comparison that should nag at anyone tempted by the growth: Amazon, Microsoft, and Google ran the same build-ahead-of-revenue playbook. They did it funded from internal cash. Oracle is, according to one analysis, spending an estimated 2.2 to 2.6 times its operating income on capex, versus roughly 1.6 times for a hyperscaler peer funded entirely from its own income.
Same model. Much weaker starting position. And a concentration risk the hyperscalers never carried.
So when Magouyrk says the new contracts required no additional capital from Oracle, read it precisely. He is telling you those specific deals were customer-financed. He is not telling you the $70 billion build behind the backlog is.
The revenue is arriving on time. That was the easy half, and the print answered it convincingly.
The cash is still going out — negative $5.4 billion this quarter, negative $23.7 billion last year, and not modeled to turn until 2029. The reassurance was engineered to keep your eyes on the first number.
Watch the second one. And watch the one name that governs it.