Oracle's AI Trade Just Got Broader — and the Backlog Finally Reaches the P&L

Around this time last year, Oracle delivered one of the most explosive earnings prints the stock market has seen. A single fiscal first quarter added roughly $455 billion of signed future business and sent shares up more than 40% in a day, briefly making founder Larry Ellison the world's richest man. That move rested almost entirely on one contract: a $300 billion, five-year cloud deal with OpenAI, the company behind ChatGPT. And it had an immediate flaw. Within days, the stock dropped 6% after an analyst warned that too much of Oracle's upcoming growth is coming from a single client: OpenAI.

That was the fear Thursday's fiscal-first-quarter report was built to answer. It answered it about as directly as a print can.

The numbers: revenue rose 30% to $19.3 billion, powered by the cloud-infrastructure business Oracle calls IaaS, which more than doubled — up 121% to $7.4 billion. Adjusted earnings of $1.92 beat the $1.74 analysts expected. The company raised its full-year forecast to at least $90 billion of revenue and $8.10 in adjusted earnings per share. And the backlog — the signed but not-yet-delivered business known formally as remaining performance obligations, or RPO — climbed to $664 billion from $638 billion at the start of fiscal 2027. Shares jumped as much as 7% in extended trading.

The difference between a promise and a machine

The distinction that matters is what the year-ago number and this-year number each represent. A year ago, Oracle sold investors on expectation: one enormous OpenAI deal sitting in the backlog. RPO is contracted future revenue; it is not income yet, and it is not cash in the bank. Greyhound Research put it plainly in a widely read piece: a backlog isn't a bank balance.

This quarter, the backlog started landing in the income statement. The 121% IaaS growth is converted revenue, not a vow. And management says the mix has broadened — recorded contracts with AI names beyond OpenAI, including xAI and Meta, plus enterprise customers and a Pentagon deal worth up to $7 billion over a decade. Chief financial officer Hilary Maxson said roughly half of the new backlog will convert into revenue over the next 36 months without requiring incremental capital. The single-customer concentration that spooked the market last fall is measurably receding. The bull case that canceled the OpenAI-only worry is that this is now diversified AI demand, not one giant bet.

The most expensive conversion in software

Here is where the honest part arrives, because converting that much contract is not free.

Oracle spent $28.5 billion on capital expenditures in the quarter — more than triple the $8.5 billion a year earlier — and guided to $90 to $95 billion for the full fiscal year. Free cash flow came in at negative $5.4 billion, versus roughly breakeven a year ago. The balance sheet carries $125 billion of debt, and S&P Global downgraded Oracle's credit rating in July. This is a machine that turns debt and equity into data centers — 850 megawatts of capacity and more than 300,000 GPUs brought online in one quarter — betting that contracted revenue outruns the spend.

The scarcity logic is real. Oracle insists AI compute demand still grows faster than supply, and a chunk of its new capacity is funded through prepaid and "bring-your-own-hardware" contracts that do not consume more of Oracle's own cash and will not hit its capex or revenue until fiscal 2028 at the earliest. If compute is the scarce complement to an economy that is suddenly abundant in AI output, an owner of contracted capacity collects scarcity rents. On one measure, that scarcity is priced in Oracle's favor: one analyst called OCI demand compounding exponentially faster than AWS or Azure, on AI compute gross margins of 30% to 40%.

But abundance-scarcity cuts both ways. Every hyperscaler is pouring money in at the same scale; if the industry's collective buildout overshoots realized demand, scarcity inverts and those fat margins compress. RPO is a contract, not a bank balance, and the war is being fought with borrowed money.

The race now deciding the stock

At roughly $160, Oracle trades near the bottom of its 52-week range and about half its $345 peak — around 19 to 21 times forward earnings, cheaper on that basis than Microsoft or Amazon. The market has already discounted a meaningful share of the "one OpenAI client" and "unfunded cash burn" fears. What Thursday's report supports is that the diversification story is real and the conversion has begun; the data no longer supports the one-customer caricature.

What it does not settle is whether conversion finishes ahead of the buildout. The decisive question is no longer whether demand exists — it demonstrably does — but whether a company borrowing $125 billion and spending nearly $100 billion a year to turn contracts into cash can do so faster than the industry's shared capex makes compute abundant. That is the race Oracle is in now, and the report just moved it a lap ahead.