Oracle's record cloud backlog shows AI infrastructure demand is still expanding, even as capex pressure rises.
Oracle has become one of the cleanest public-market tests of the AI infrastructure cycle. Its latest earnings show that demand for cloud compute is still climbing, but they also show why investors are scrutinizing the cost of building that capacity.
The company reported fiscal fourth-quarter revenue of $19.2 billion, up 21 percent from a year earlier. Cloud revenue rose 47 percent to $9.9 billion, while cloud infrastructure revenue jumped 93 percent to $5.8 billion. Those numbers are unusually strong for a company that was long viewed mainly as an enterprise software incumbent.
The more striking figure is backlog. Oracle said remaining performance obligations rose to $638 billion, up 363 percent year over year and $85 billion from the prior quarter. The company said most of the recent increase came from large-scale AI contracts, including arrangements where customers prepaid for GPUs or supplied hardware to Oracle.
That structure helps explain both the excitement and the concern. AI labs and enterprise customers need enormous blocks of compute for training, inference and data-heavy applications. Oracle is trying to convert that demand into a durable cloud infrastructure business. But data centers require capital before revenue fully arrives.
Oracle said free cash flow was negative $23.7 billion for fiscal 2026 as it invested to support cloud infrastructure growth. It also expects to raise about $40 billion in fiscal 2027 through debt and equity financing, including a previously announced $20 billion at-the-market equity issuance.
The result is a sharper version of the broader AI capex debate. Bulls see contracted demand, cloud revenue growth and customer-funded hardware as evidence that the buildout is real. Skeptics worry that the industry is racing ahead of monetization and that only some providers will earn attractive returns on the capital being committed.
Oracle's numbers do not settle that debate. They do show that the AI compute boom is still expanding, and that the next phase will be judged less by model demos than by power, financing, margins and the ability to turn backlog into reliable capacity.
Image source: i.ibb.co