Oracle's "Non-Linear" Capex Isn't a Cost Problem. It's a Capital Problem.

Oracle reported its largest quarter ever on Thursday — revenue up 30%, adjusted earnings up 30%, cloud infrastructure revenue up more than 120% — and the stock still fell, and is down more than half from last September's peak. The reason investors keep rejecting record results is the same one they have cited for four straight quarters: the money going out the door to build data centers.

That worry is real, but it is aimed at the wrong target. Fixating on the size of the buildout misses what actually changed on this call. It is not the full-year capex plan of $90 billion to $95 billion, which has been on the table since spring. It is the sentence the finance chief used to describe it: capex will not be linear throughout the year.

That phrase is a pacing disclosure, and reading it carefully explains more about Oracle's current position than the revenue beat does.

A $28 billion quarter, and the hedge

Oracle spent $28.5 billion on capital expenditures in the fiscal first quarter — roughly a third of its entire annual budget in a single quarter. Naively annualized, that run-rate would exhaust the yearly range in three quarters. "Not linear" is the pre-emptive answer to that arithmetic: the spend is lumpy, tied to when contracted data center capacity comes online and when batches of GPUs ship, so one quarter's pace is not the run-rate.

The disclosure carries a quieter claim as well: that the annual figure is a cap, not a trajectory. That deserves scrutiny, because this is the same management that guided fiscal 2026 capex to about $50 billion and spent $55.7 billion.

The number that isn't entirely Oracle's

The more important detail sits further down in the cash flow. Gross capex of $28.5 billion cost Oracle only about $18 billion in cash, because customers prepaid $11.4 billion against those same AI contracts, with financing components attached.

This is the mechanism the headline figures obscure. Oracle is not funding the buildout alone. In fiscal 2026 it raised $43 billion of debt and $5 billion of equity, and customers prepaid or supplied $75 billion worth of hardware into the backlog. This year it expects about $25 billion in customer reimbursements. The gross number flatters the fear; the net cash Oracle actually has to come up with is smaller.

But only smaller. This is not free money, and it is not self-funding yet.

Capex now roughly equals revenue

Scaling the numbers shows how far Oracle has traveled from a software company. It spent $55.7 billion on capex in fiscal 2026 on $67.4 billion of revenue — 83 cents of capital for every dollar of sales. This year it has guided capex of $90 billion to $95 billion against revenue of at least $90 billion, putting capital spending at roughly 100% of revenue.

No peer runs this hot. Microsoft, Alphabet, Amazon and Meta spend heavily on AI but generate positive free cash flow and can self-fund. Oracle's free cash flow is deeply negative — minus $23.7 billion in fiscal 2026 — so the buildout is funded by three things: customer prepayments, and roughly $40 billion of new debt and equity this year, including a $20 billion at-the-market share sale that dilutes holders at a depressed price.

The constraint has migrated. It is no longer about whether Oracle can build capacity; it delivered more than 300,000 GPUs in a quarter and added 850 megawatts. The question now is whether the capital keeps arriving faster than the cash does.

The single name behind half the backlog

That is where concentration enters. Remaining performance obligations — contracted but not yet delivered revenue — reached $664 billion, and roughly half of that is attributed to one customer, OpenAI, which does not yet earn a profit. S&P Global Ratings downgraded Oracle in July to one notch above speculative grade, naming OpenAI concentration as a key credit risk.

The logic of the whole buildout rests on a single assumption: that today's contracted demand converts into revenue and cash. Management expects about $77 billion of the backlog, roughly 12%, to be recognized within the next year. If that conversion proceeds on schedule, the capex peak eventually becomes self-amortizing, as the backlog turns into operating cash that defrays the next buildout. If the big prepayer stumbles for a quarter, or a tranche slips, Oracle holds capacity built for a customer that has gone quiet, and a capital stack that still needs servicing.

So the honest reading of "non-linear capex" is neither reassurance nor alarm. It is a statement about pacing, layered on a structure where the real risk was never the quarterly spend. It is whether the backlog converts into cash faster than the debt and dilution required to build it — a condition that rests on the timing of a single unprofitable customer. That is the number to watch, not the quarter's buildout.