Oracle’s 93% Cloud Growth Is Real. The More Important Story Is Who’s Paying for the Hardware
Oracle’s cloud infrastructure revenue grew 93% in its fiscal fourth quarter. The company burned $23.7 billion in free cash flow over the same fiscal year. The difference is not a contradiction. It is the business model.
Large AI customers have prepaid for or directly supplied $75 billion worth of GPU hardware tied to Oracle’s long-term infrastructure contracts. Oracle carries the buildings, the power, and the operations. Its customers carry part of the chips.
The $75 Billion Number Is Not What It First Appears
When Oracle disclosed the $75 billion figure in its fiscal Q4 2026 earnings release on June 10, 2026, early coverage described it as “$75 billion in GPU prepayments.” The actual disclosure is more specific and more interesting.
The $75 billion combines two distinct arrangements. In the first, customers pay Oracle in advance so Oracle can purchase GPUs on their behalf. In the second, customers buy the hardware themselves and supply it to Oracle-operated data centers. Oracle then provides the facilities, power, cooling, networking, software, and operations.
Several things the $75 billion is not: quarterly revenue, annual recurring revenue, the total value of the associated cloud agreements, or necessarily cash Oracle has already received. A portion of the hardware may remain economically or legally associated with the customer rather than with Oracle.
Oracle has not disclosed how much of the $75 billion falls into each category, which customers are involved, or how much cash has changed hands. The details matter for margin analysis. The strategic picture is clear regardless: Oracle has persuaded some of the world’s largest AI companies to help fund the physical infrastructure behind their future cloud consumption.
Oracle’s Infrastructure Business Nearly Doubled
The headline figure from Oracle’s fiscal Q4 2026 results is 93% growth in Oracle Cloud Infrastructure revenue, which reached $5.8 billion for the quarter. Total cloud revenue hit $9.9 billion, up 47% year over year. Total company revenue was $19.2 billion, up 21%.
For the full fiscal year 2026, Oracle reported $67.4 billion in total revenue, up 17%, with cloud revenue of $34.0 billion, up 39%. OCI infrastructure revenue for the full year reached $18.1 billion, up 77%.
The scale of customer commitment is visible in Oracle’s remaining performance obligations, the contracted revenue the company has not yet recognized. At the end of Q4, the figure stood at $638 billion, up from $553 billion at the end of Q3. The $85 billion sequential increase in a single quarter reflects the pace at which large AI labs are committing to long-term cloud infrastructure capacity.
AI Customers Are Becoming Infrastructure Financiers
The two models behind the $75 billion figure represent a meaningful departure from standard cloud procurement.
Under the prepayment arrangement, a customer transfers capital to Oracle before a facility becomes operational. Oracle orders and installs the hardware. The customer reserves the compute capacity it expects to need. For Oracle, the arrangement reduces upfront financing requirements and tightens the link between infrastructure spending and contracted demand. For the customer, prepayment can accelerate access to large AI clusters and provide insulation from supply shortages.
Under the customer-supplied model, the customer purchases GPUs directly and delivers them to an Oracle-operated facility. Oracle contributes the data center, power, networking, and software layer. The arrangement more closely resembles a managed infrastructure service than a conventional public cloud purchase.
Oracle has publicly associated its infrastructure expansion with customers including OpenAI, Meta, and NVIDIA, and the company participates in the Stargate initiative. It has not disclosed which customers account for any portion of the $75 billion.
The Model Reduces Capital Needs but Does Not Remove Them
The customer-financing model matters precisely because Oracle’s capital requirements are enormous. The company spent approximately $55.66 billion on capital expenditure in fiscal 2026, above its previous $50 billion target. Free cash flow for the year was negative $23.7 billion. Oracle funded the gap with $43 billion in debt financing and $5 billion in equity.
For fiscal 2027, Oracle expects approximately $70 billion in company-funded capital expenditure, with an additional $20 billion to $25 billion in customer-reimbursed expenditure. The company plans to raise approximately $40 billion through debt and equity to support the program.
The $75 billion in customer prepayments and customer-supplied hardware helps Oracle fund the hardware component of AI infrastructure without placing the full cost on its balance sheet. The buildings, power connections, transformers, cooling equipment, and high-speed networking still require Oracle’s capital. The shortage constraining AI infrastructure today extends well beyond GPU supply: it includes electrical grid capacity, gas turbines, land, construction labor, and permitting timelines.
Customer-supplied hardware may improve Oracle’s return on invested capital. It does not automatically raise margins. Customer reimbursements may pass through at limited margin, Oracle may have offered favorable pricing to secure long-term commitments, and data-center and power costs remain substantial regardless of who owns the GPUs.
Oracle Is Growing Faster From a Smaller Base
Oracle’s 93% OCI growth rate exceeds the most recently reported figures from its larger infrastructure competitors, but the comparison requires context.
AWS reported $37.6 billion in revenue for Q1 2026, up 28% year over year, with operating income of $14.16 billion. Amazon has said AI-related revenue within AWS reached a $15 billion annualized run rate. Google Cloud reported revenue exceeding $20 billion in Q1 2026, up 63%, with a backlog above $460 billion. Microsoft reported Azure growth in the high-30% range during fiscal 2026 but does not publish standalone Azure revenue figures.
Oracle’s $5.8 billion quarterly IaaS figure remains substantially smaller in absolute terms than AWS or Google Cloud. The 93% growth rate reflects a lower starting base rather than a reversal of the competitive gap. The defensible argument is not that Oracle has overtaken the hyperscalers. It is that Oracle has built a real position in very large AI clusters and has done so at a speed the larger incumbents are watching closely.
Backlog Creates Visibility and Concentration Risk
The $638 billion remaining performance obligations figure is an asset and a risk at the same time.
Oracle expects to recognize approximately 12% of RPO during the following 12 months, roughly $76.6 billion, and another 34% during the subsequent two years, approximately $216.9 billion. The remaining 54%, approximately $344.5 billion, falls outside a three-year window. Revenue recognition depends on Oracle delivering operational infrastructure on schedule: completed buildings, live power connections, and functioning hardware.
The delivery dependency is a concentration risk as well as a construction challenge. A relatively small number of AI companies likely accounts for a substantial share of Oracle’s backlog. The loss, delay, or financial difficulty of a single large customer would have a disproportionate impact on Oracle’s revenue trajectory. AI companies themselves depend on continued fundraising and uncertain future revenue, and long-term contracts do not eliminate counterparty risk.
The Cloud Contract Is Becoming a Construction Contract
Oracle’s customer-financing model reflects a broader structural shift in how large-scale AI infrastructure gets built and who pays for it.
Hyperscale AI deployment now requires coordinating GPU delivery, power procurement, construction timelines, cooling capacity, and network commissioning across campuses that may cost several billion dollars each. The timelines for electrical interconnection agreements and onsite generation exceed typical software procurement cycles by years. Customers wanting guaranteed access to capacity at scale are increasingly drawn into those timelines and costs.
Oracle has made that participation explicit and contractual. Customers that prepay or supply hardware are co-investors in the physical infrastructure their AI systems will run on. The arrangement accelerates Oracle’s construction program by aligning customer capital with its build schedule, and it ties large customers more closely to Oracle Cloud Infrastructure over contract periods that may span five years or more.
Oracle did not find free infrastructure. It found a way to share the cost of building it. The size of its $638 billion backlog suggests the model is working, at least for now. The harder test comes when those contracts require delivery.
