Wouldn’t it be funny if Larry Ellison, who has become the elder statesman of the datacenter, had the last laugh on the cloud builders and model builders by beating them at their own game?
That these upstarts invent this GenAI technology – much as IBM invented the relational database and SQL queries, BEA Systems invented web application serving, and SAP, PeopleSoft, Siebel Systems, and dozens of others invented application suites – but it is Oracle that might build the most infrastructure for GenAI and that sells the most actual GenAI functionality through its applications is, well, funny.
Oracle is beginning to look like a much safer bet as a massive-scale infrastructure supplier than it might have only three years ago.
There are many reasons why this has become true, and they are converging to give Oracle, the company that Ellison co-founded with Bob Miner and Ed Oates five decades ago to beat IBM to market with the relational database in the datacenter, some pretty big advantages. Buying Sun Microsystems back in 2009 gave Oracle system hardware and operating system chops that it was lacking to underpin its obvious expertise in middleware, databases, and application software.
Oracle was late to the cloud, but in the past several years it has more than made up for it.
Like Google, Oracle started with platform-level cloud services where it was basically offering APIs for storage and compute to customers – and they didn’t want it. So in 2015, three years after launching Oracle Cloud, Ellison put a stake in the ground outlining how it could – and would – compete against Amazon Web Services, which seemed unassailable in cloud, and Microsoft Azure, which was the real contender due to its vast Windows Server and SQL Server bases as well as its application stacks. A year later, the company unveiled infrastructure as a service and rebranded its utility as Oracle Cloud Infrastructure, and it has been growing steadily since then.
Oracle is a massive enterprise software engine, and now with AI infusing all of its development processes, it is partnering with GenAI model builders to help add agentic AI to its application suites and, as Ellison explained in a call with Wall Street analysts going over the financial results for Oracle’s first quarter of fiscal 2026, to actually create new applications that are really just collections of agents.
Ellison elaborated its AI advantage on the call, which featured a massive uplift in the forecast for OCI out for the next five years based on current bookings out that far. We quote Ellison at length about the AI inference opportunity for Oracle because sometimes you have to:
“AI inferencing will be used to run robotic factories, robotic cars, robotic greenhouses, biomolecular simulations for drug design, interpreting medical diagnostic images and laboratory results, automating laboratories, placing bets in financial markets, automating legal processes, automating financial processes, automating sales processes. AI is going to generate the computer programs, called AI agents, that will automate your sales and marketing processes. Let me repeat that. AI is going to automatically write the computer programs that will then automate your sales processes and your legal processes and everything else in your factories and so on.”
“Think about it: It’s AI inferencing that will change everything. Oracle is aggressively pursuing and we’re not doing badly in the AI training market, by the way. Oracle is aggressively pursuing the inferencing market as well as the AI training market. We think we are in a pretty good position to be a winner in the inferencing market because Oracle is by far the world’s largest custodian of high-value, private, enterprise data.”
“With the introduction of our new AI database, we added a very important new way for you to store your data in our database. You can vectorize it. And by vectorizing all your data, all your data can be understood by AI models. Then we made it very easy for our customers to directly connect all their databases, all their new Oracle AI databases and OCI cloud storage to the world’s most advanced AI reasoning models – ChatGPT, Gemini, Grok, Llama, all of which are uniquely available in the Oracle Cloud. After you vectorize your data and link it to an LLM – the LLM of your choice – you can then ask any question you can think of.”
“For example, how will the latest tariffs impact next quarter’s revenue and profit? You ask that question, the large language model will then apply advanced reasoning to the combination of your private enterprise data plus publicly available data. You get answers to important questions without ever compromising the safety and security of your private data. Again, I’d like you to think about this for a moment. A lot of companies are saying, ‘We’re big into AI because we’re writing agents.’ Well, guess what, we are writing a bunch of agents, too. But when they introduced ChatGPT almost three years ago, what you got to do is have a conversation and ask questions. You weren’t automating some process with an agent. You could ask whatever question you wanted to ask and get a well-reasoned answer with all of the latest and best information and high-quality reason go along with it.”
“Who’s offering that to customers? We’ll be the first when we deliver it and demonstrate it at AI World next month.”
Later in the call, Ellison said that Oracle had built “the most advanced application generator of any company,” and said the company was no longer building applications as such, but generating them through its AI stack, and moreover, that it does not charge separately for its AI stack because the new applications are themselves AI. The applications are better, Ellison added, and as such, they hope to sell more of them and make more money that way from AI. This is a better business plan than OpenAI, which looks like it will be Oracle’s biggest cloud infrastructure customer, has put together to make money from AI.
Selling AI infrastructure for training and inference has taken off like a rocket for Oracle, which of course was the big news while we were unironically attending the AI Infra Summit in Santa Clara last week.
Safra Catz, chief executive officer at Oracle for many years as Ellison, who is very healthy looking 81, is chairman and chief technology officer, jumped right in, telling Wall Street that the company’s remaining performance objectives (what we would call a revenue backlog) ended the August quarter at $455.3 billion, up by a factor of 4.6X year on year and up 2.3X from the fourth quarter of fiscal 2025 that ended in May. Catz walked through a new financial segmentation and then said for Q2, she expected for RPO to grow and exceed $500 billion, and then laid out very precise revenue figures for OCI: $18 billion in fiscal 2026, $32 billion in fiscal 2027, $73 billion in fiscal 2028, $114 billion in fiscal 2029, and $144 billion in fiscal 2030.
We have never seen such a forecast, in either magnitude of years or precision of numbers, and it is indicative of the faith Oracle has in the customers behind that huge backlog. Or, the world has gone completely bonkers and there are no consequences for hyperbole anymore.
And even if Ellison doesn’t get the last laugh, for the next several years, it looks like Oracle will be laughing all the way to the bank. . . . And if these deals don’t come to pass, Ellison’s 40 percent stake in the company sure did skyrocket on the news.
To put that forecast into some sort of perspective, we plotted out Oracle’s overall cloud revenues, its specific OCI revenue within that, and its capital expenses between fiscal 2013 and fiscal 2026, which is estimated. We do not have a good sense of OCI revenues prior to fiscal 2021, but it was clearly pretty small, relatively speaking. But it has taken off nicely as Oracle has gotten AWS, Microsoft Azure, and Google Cloud to take its Exadata iron when they sell cloudy Oracle databases and as Oracle has become a player in GPU infrastructure.
In fiscal 2025, Oracle spent almost as much on capital expenses as it made on overall cloud revenues, which is an astounding thing, and it looks like in fiscal 2026, these two figures will cross over. Under normal circumstances, these facts would put tremendous downward pressure on any company’s stock. But Oracle put out a very long forecast, and it has a very profitable enterprise software business that can help it make continuing investments in OCI. Oracle is rich enough to build iron for AI model makers like OpenAI, and it is being contracted to do so.
Just for fun, we took Catz’s OCI projections out to fiscal 2030 and then did our best to model what Oracle’s overall cloud revenues would be and also what its capital expenses might be. Take a look:
Yes, fiscal 2026 is a very heavy spending year for Oracle, but as Catz reminded Wall Street, Oracle does not actually build datacenters, but rather just the compute, storage, and networking that goes into a datacenter that the customers either build themselves or rent from co-los.
Eventually, we think that Oracle’s capex will moderate and its OCI revenues will grow faster and be larger, and the gap between the two is profits. The gap between overall cloud revenues and capex is also profits, and we think the red line will follow the black one up as it must given the hundreds of thousands of application, middleware, and database customers that Oracle has worldwide.
The upshot is that Oracle’s cloud business, which includes cloud applications and cloud infrastructure, looks like it will easily break through $100 billion in fiscal 2029, and we think it will in fact be $23 billion for SaaS and PaaS services across all of Oracle’s software paired with that $114 billion in projected OCI revenues. That is 1.65X growth between fiscal 2021 and fiscal 2029 for SaaS and PaaS cloud services against 11X growth for OCI.
The Oracle we will come to know is very different from the Oracle that we have always known.
Back in the present, let’s go over the numbers.
In the August quarter, Oracle had $14.93 billion in sales, up 12.2 percent, with operating income up 7.2 percent to $4.28 billion and net income off a smidgen to $2.93 billion. Oracle ended the quarter with just a tad over $11 billion in cash and equivalents, which is obviously a lot lower than it has been in prior years but understandable given all of the OCI investments in the past decade.
Here is a table showing the new financial segmentation since Q1 F2025:
Oracle lumps together cloud applications and cloud infrastructure so it doesn’t have to show the margins for its cloud infrastructure business, which is no doubt a lot lower than for its cloud infrastructure. Margins for the systems that Oracle sells to customers – simply Hardware – has pretty decent operating margins, which shows the benefits of specialization and not trying to compete with other OEMs and the ODMs of the world. This is not a big OEM hardware business, of course, but we are not seeing all of it. Cloud infrastructure rentals also drive hardware revenues for Oracle.
Since the beginning of fiscal 2021, Oracle has put comments in its financial releases that show its cloud revenues and a breakdown of IaaS and SaaS revenues to two significant digits. This is not highly precise data, but it is illustrative:
Oracle no longer really has a PaaS business. Or, the dividing lines are smudged. Is a database infrastructure or higher-level software? You could argue either way. We are not sure how Oracle carves it up, to be honest, but we will try to find out.