Oracle 云业务开始回报：Q1 F2027 营收 191.8 亿美元，IaaS 翻倍至 74 亿

CLOUD

In its latest quarter, Oracle set record revenue levels and, importantly for the first time in its five decade history, its first quarter of one fiscal year was larger than the fourth quarter of the prior fiscal year. This is bucking the normal cycle where salespeople try to cram as many deals into Q4 to get a big bonus, which understandably makes Q1 naturally a weak one.

Granted, it was only eight-tenths of a percent growth sequentially, so it was not huge. But it is a different pattern, and suggests that Oracle, which had a revenue backlog of $664 billion as the quarter ended, is going to see its cloud infrastructure business go superlinear if not exponential. And that is being driven by backlog converting into sales.

“We would expect acceleration to continue in the remainder of fiscal 2027 as we convert more RPO into revenues,” Hillary Maxson, Oracle’s chief financial officer, explain on a call with Wall Street analysts going over the numbers for Q1 F2027. “We now expect around half of our RPO to convert into sales over the next 36 months.” If you do the average on that, it works out to $27.7 billion per quarter. It will be smaller in fiscal 2027 and larger in fiscal 2029, of course. Exactly how much is hard to predict from the outside of Oracle, of course. And then the other half of the backlog will still be there out beyond fiscal 2029.

In the quarter, Oracle posted revenues of $19.18 billion, up 29.6 percent and up a smidgen sequentially (but not down). Operating income rose nearly twice as fast, up 57.3 percent to $6.73 billion, and net income grew even faster, up 62.6 percent to $4.76 billion.

Under any normal circumstances, with Oracle approaching a net income to revenue ratio that predates its heavy cloud investments, such numbers would make Wall Street gaga. But they are worried about a future where Oracle plans to spend $90 billion to $95 billion in capital expenses for fiscal 2027. Oracle just raised $19.9 billion floating stock, and thanks to $11.4 billion prepayments largely from cloud customers, it had a cash flow of $23 billion in the quarter. But it also blew $28 billion on capex, which makes its cash flow a negative $5 billion.

Here is the monster table breaking down revenues and operating income for Q1 F2027 for Oracle and going back to Q1 F2026:

The Oracle Cloud division made $11.61 billion in sales, up 61.5 percent and up 17.1 percent sequentially. Software license sales took their typical Q1 dive, down 65.2 percent sequentially and down 14.5 percent year on year as customers continue to shift to cloud-based subscriptions from perpetual licenses plus support contracts. The software support business, which is basically an annuity that Oracle cashes in on every quarter, had $4.9 billion in sales, down 1.2 percent.

Hardware sales – yes, customers still buy Oracle infrastructure rather than renting it – was up 15.5 percent to $774 million, but was down 16.2 percent sequentially. In the year ago period, hardware sales were actually up 6.3 sequentially to $670 million. Operating income for hardware is under pressure as component prices for systems are going haywire thanks to the GenAI boom, where demand is outstripping supply on nearly all semiconductor fronts, and have hit a low of 63.7 percent of revenue compared to an average north of 73 percent in the two years before the component price hikes hit. Interestingly, the operating margins are under the same pressure in the Cloud And Software group, no doubt caused by component prices for the iron Oracle buys for its cloud as well as increasing expenses for all components of a datacenter buildout.

Services, which includes implementation of Oracle’s myriad applications in its Fusion and NetSuite portfolios as well as many legacy application stacks that don’t change much and for which Oracle can rightfully charge an arm and a leg to support because customers have not moved ahead, rose by 4.8 percent to $1.41 billion.

The exciting bit for the Oracle top brass is that Oracle’s cloud infrastructure (IaaS) revenues rose by a factor of 2X to $7.4 billion in the quarter and was up 27.6 percent sequentially. By comparison, Oracle’s SaaS software revenues only rose by 9.2 percent to $4.2 billion.

Here is the drilldown Oracle presented to Wall Street on its IaaS and SaaS revenues in the Cloud division:

Oracle’s cloud database business – which one could argue should be considered SaaS because it is software, but then again it is systems software running on Exadata hardware that has been highly tuned for it – was up 26 percent to $900 million. Within this segment, the multi-cloud implementation of the Oracle database sold as an infrastructure service rose by 4.5X year on year. It is very likely a small part of the overall database slice of IaaS sales.

Sales of CPU and GPU capacity comprise around 88 percent of Oracle’s cloud IaaS sales and generated $6.5 billion in sales, up 2.5X year on year. This is the conversion of AI backlog to running workloads. Oracle added 850 megawatts of its own infrastructure capacity with more than 300,000 GPUs and untold CPUs to rent out by the hour during fiscal Q1, which is 3X the capacity Oracle delivered in Q4 F2026 (around 280 megawatts). Oracle added 1.16 gigawatts of capacity last year, with 400 megawatts coming in Q3 F2026, leaving 480 megawatts of capacity being installed in 1H F2026.

For comparison, the OpenAI Stargate buildout, where Oracle is installing the infrastructure for the AI model builder, in Abilene and Shackleford in Texas (3.2 gigawatts) and Dona Ana, New Mexico (2.2 gigawatts) and Port Washington, Wisconsin (1.3 gigawatts) and Saline, Michigan (1.4 gigawatts) are supposed to added up to 8.1 gigawatts. Oracle says it has secured 10 gigawatts of power for three years through fiscal 2029. Those are datacenter power numbers, so the infrastructure power draw is probably on the order of 70 percent of that, or about 5.7 gigawatts. If that is comprised of Nvidia systems, that would cost $227 billion to deliver just the infrastructure at the expected $40 billion per gigawatt cost that Nvidia chief executive officer Jensen Huang has talked about. If it is a mix of homegrown OpenAI Jalapeno accelerators (for inference) and Nvidia GPUs (for training), it probably blends to around $30 billion per gigawatt, then it will cost $170 billion for the hardware.

Clay Magouyrk, co-chief executive officer at Oracle and the person in charge of the Oracle Cloud Infrastructure business, told Wall Street that 131,000 GPUs (almost certainly “Blackwell” B200s and B300s from Nvidia) went into the Abilene datacenter for OpenAI in the quarter, and 69,000 went in during Q4 F2026. At this point, OpenAI has taken delivery of 618 megawatts, or 75 percent of the total capacity, of the infrastructure going into Abilene from Oracle. The new GPT-6 Astra model was trained on this iron in Abilene. The Shackleford datacenter is being fleshed out with “Vera-Rubin” systems from Nvidia, and OpenAI will get its hands on this initial capacity sometime during Q2 F2027.

For the moment, the GPU economics are working in the favor of Oracle as GPU and XPU demand far outstripped supply. (How much is unclear.) The majority of the GPUs in Oracle’s fleet – which has to number 1 million or more GPUs by now – are four years old or older, and yet they are commanding higher and higher prices because compute need is bigger than compute capacity. It is not hard to see these devices being in use for three, four, or even more years into the future – and they probably pay for themselves in three years. Across all of the GPUs that came up for renewal in Q1, the portfolio of renewals priced out at 20 percent higher than their original contracts. GPU utilization was at 97.9 percent in Q1, which is IBM mainframe-class utilization.

I will end with some interesting notes on Oracle’s own use of AI.

Mike Sicilia, co-chief executive office and the person who runs the software side of the Oracle house, said on the call that Oracle would be announcing a new agentic AI accelerator at its AI World event next month that would use GenAI to deploy Oracle’s applications in the field at a much faster pace than can be done with human implementers today – and at a much lower cost. Implementation costs for ERP suites can be significant

“Working alongside Oracle and customer teams, AI agents will automate and orchestrate implementation at an unprecedented scale, compressing SaaS deployments from years to months and months to weeks,” said Sicilia. “It is really the power of these things together that reinforce my belief that the growth of our applications business is only going up from here.”

Customers running their Oracle applications in the OCI infrastructure tapped embedded AI capabilities in their applications 150 million times in Q1, which is a 42 percent sequential growth, and AI agents were launched more than 3.5 million times, which was nearly double sequentially from Q4 F2026. Customers, says Sicilia, have over 2,300 agents in production on their Oracle applications. In the high-end Fusion ERP suites alone, customers burned 900 billion tokens.

“AI changes this dynamic,” Sicilia explained. “Rather than asking every employee to navigate and execute a process exactly as a system expects, AI agents can perform tasks using the organization's established workflows and business rules. Employees then shift to overseeing agents, resolving exceptions, and applying human judgment where it matters most.”

SPONSORED FEATURE: HPE and NVIDIA help organizations apply accelerated computing, software, enterprise infrastructure, networking, control plane, and services to the workloads and business outcomes that matter most.

October 1, 2026

September 30, 2026

AMD Buys World Labs For $8.2 Billion To Get AI Model Cred