$664 Billion in Signed Contracts: Oracle's Q1 Makes It the AI Boom's Landlord

The most consequential number in this quarter’s AI economy was not printed on any model card. It appeared in Oracle’s fiscal Q1 2027 earnings release on September 10: remaining performance obligations of $664 billion, up $209 billion in a single year. That is contracted, signed, mostly prepaid future revenue — customers who have locked in AI compute capacity years ahead, at full commitment, because they are terrified of being the company that cannot get GPUs when it needs them.

Oracle’s quarter, on its face, was a beat across every line. Total revenue rose 30% to $19.3 billion. Cloud revenue (IaaS + SaaS) grew 62% to $11.6 billion, driven by cloud infrastructure (IaaS) surging 121% to $7.4 billion — a triple-digit growth rate at multi-billion scale that no other company in the industry is posting. GAAP operating income rose 57% to $6.7 billion; non-GAAP EPS climbed 30% to $1.92. The stock closed up nearly 7% at $163.29 on 2.4 times normal volume.

But the numbers underneath the headline tell a stranger and more important story — one about what happens when demand for AI compute structurally outruns any company’s ability to build it.

The backlog is the business

Remaining performance obligations — RPO, in finance jargon — represents revenue Oracle has contractually locked in but not yet delivered. The Q1 figure of $664 billion grew by $209 billion year-over-year, after Oracle booked more than $30 billion in new AI cloud contracts during the quarter alone.

To put that in perspective: Oracle’s entire annual revenue for fiscal 2026 was roughly $71 billion. The backlog now represents more than nine years of current-scale revenue. The company confirmed in the release that “customer demand for AI Cloud Training and Inferencing Services continues to grow faster than supply” — corporate-speak for a genuinely extraordinary situation: the constraint on Oracle’s growth is not sales, not competition, and not pricing. It is whether concrete can be poured and racks energized fast enough.

The mechanics of how this backlog is funded deserve attention. Oracle’s earnings tables show the role of customer prepayments with significant financing components — AI customers, led by OpenAI’s multi-hundred-billion-dollar compute commitments, effectively paying up front for capacity that doesn’t exist yet. Combined with bring-your-own-hardware arrangements, this is a demand-pull model in which Oracle’s customers are financing the very factories that will serve them. It inverts the usual logic of cloud capital expenditure: rather than building capacity and hoping demand shows up, Oracle is taking delivery of demand it cannot yet serve.

300,000 GPUs, 850 megawatts, 97.9% utilization

The physical footprint of that demand showed up in three operational metrics. Since the end of Q4, Oracle delivered more than 300,000 GPUs to AI cloud customers — nearly triple the capacity delivered in the prior quarter. It added 850 megawatts of new data center capacity in a single quarter, roughly the output of a large nuclear reactor unit. And its AI infrastructure ran at 97.9% utilization — effectively full.

That utilization number is the one infrastructure analysts keep circling. A conventional enterprise cloud aims to run hot at maybe 70–80% utilization, holding spare capacity for burst workloads. 97.9% means Oracle has no headroom at all: every GPU it owns is earning, and any new customer demand has nowhere to go except into the backlog. Tripling GPU deliveries in one quarter and still growing RPO by $209 billion over the year is the clearest quantitative statement yet that AI compute supply is nowhere near demand — the same crunch that this week had OpenAI freezing new ChatGPT Pro signups on the consumer side and Microsoft mapping a 38-gigawatt data center roadmap on the enterprise side.

The balance sheet bet

Here is where the story stops being a victory lap. Oracle’s Q1 operating cash flow was a record $23 billion, up 184% — yet free cash flow was negative $5 billion, because quarterly capital expenditures ran at roughly $28.5 billion, triple the prior-year pace. To bridge the gap, Oracle completed the sale of $20 billion of common stock through an At-the-Market (ATM) equity program during the quarter, diluting shareholders to fund the buildout. Management guided FY2027 revenue to “at least $90 billion” with non-GAAP EPS of $8.10, and Q2 revenue growth of 30–34%.

Notably, the company stated that the structuring of the new AI contracts means there is no incremental impact on its plans to raise capital — meaning the massive new bookings were designed to be largely self-funded through prepayments. Watch that sentence in future quarters; it is the hinge of the whole model.

The credit side is where the tension concentrates. S&P downgraded Oracle’s long-term rating from BBB to BBB- in July, explicitly citing OpenAI as a “key credit risk” — the concentrated exposure behind a large share of that $664 billion. Oracle is borrowing heavily, issuing equity, and running negative free cash flow to build capacity whose payback depends on AI training and inference demand remaining this hot for years. If that demand softens, take-or-pay contracts with thinly capitalized AI labs become a credit question rather than a growth story. If it holds, Oracle has locked in the cheapest capacity position in the industry.

The quiet software decline

One more line item matters: software revenues fell 3% to $5.5 billion, which Oracle framed positively as customers “continuing their migration from on-premises software to the Cloud.” Maybe so. But it marks the completion of a remarkable identity transformation — the company that spent four decades selling database licenses is now, in financial substance, a leveraged infrastructure landlord whose growth engine, margin structure, and risk profile all belong to the AI capex cycle. Cloud applications (SaaS) grew a modest 10% to $4.2 billion, hardware ticked up 15%, and services rose 5% — rounding errors next to the IaaS machine.

The quarter also carried Larry Ellison’s product agenda: announcements of an AI Data Platform that automatically generates enterprise ontologies — an automated answer to what Palantir does by hand — and a “100% agentic” AI health care management and EHR system built from specialty-specific agents. But these are forward-looking demos set against a present-tense reality that is almost entirely about megawatts and GPU counts.

What to watch

Three signals will define whether the $664 billion converts into the greatest revenue backlog in software history or the industry’s most expensive stranded asset. First, the RPO burn-down rate: if cloud infrastructure revenue keeps compounding at triple digits while RPO keeps growing, the model works; if RPO stalls while IaaS growth decelerates, the market will start pricing the backlog as a funding problem. Second, OpenAI’s credit trajectory — the labs’ ability to raise capital against future model revenue is now, bizarrely, a line item in Oracle’s risk profile, and the push to get OpenAI investment-grade rated after its IPO is directly a refinancing story for Oracle. Third, the METR-era disclosure environment aside, plain capacity physics: 850 megawatts a quarter is a blistering pace, and Oracle says it is bringing on record levels of new megawatts — but the backlog implies it would need to roughly double that pace to catch up with signed demand.

For now, the market’s verdict is unambiguous. Oracle has become the fourth hyperscaler — and the only one whose growth rate accelerates as it gets bigger, because it is selling the one thing in the AI economy that cannot be manufactured fast enough: turnkey AI compute, contracted years in advance, paid for before it exists.