Oracle shifts into AI data centres
- Oracle’s AI push now centers on cloud infrastructure, with Oracle and OpenAI developing Stargate capacity and Oracle marketing large GPU clusters and data-center buildouts. - Oracle said in March 2026 that remaining performance obligations reached $553 billion, while OCI infrastructure revenue rose 84% year over year. - OpenAI said Oracle and partners were expanding Stargate sites, with additional U.S. data-center capacity and named locations disclosed in company updates.
Oracle’s recent filings, product pages and partner announcements show a company leaning much harder on physical AI infrastructure than on the software business that defined it for decades. Oracle is marketing GPU-heavy superclusters, signing multibillion-dollar cloud commitments and helping build Stargate-linked data-center capacity with OpenAI. Those moves have fed a broader industry argument that Oracle is no longer selling only databases and applications, but bundled access to land, power, chips and guaranteed compute. ### When did Oracle start talking like an infrastructure company? Oracle made that language explicit in September 2024, when it said Oracle Cloud Infrastructure was taking orders for AI supercomputers with up to 131,072 Nvidia Blackwell GPUs. Mahesh Thiagarajan, executive vice president of Oracle Cloud Infrastructure, said at the time that Oracle had “one of the broadest AI infrastructure offerings” for demanding workloads. (oracle.com) By October 2025, Oracle was describing its Zettascale10 systems as being housed in “large gigawatt data center campuses” optimized for dense AI training loads. Oracle said that architecture was being deployed with OpenAI at the Stargate site in Abilene, Texas. ### How closely is Oracle tied to Stargate? OpenAI named Oracle as one of Stargate’s initial equity funders when it announced the project, alongside SoftBank and MGX. (oracle.com) OpenAI said SoftBank had financial responsibility, OpenAI had operational responsibility, and Oracle was part of the initial funding group. OpenAI said in July 2025 that it and Oracle had entered an agreement to develop 4.5 gigawatts of additional Stargate data-center capacity in the United States. (oracle.com) OpenAI said that, together with the Abilene site, the partnership would bring Stargate capacity under development to more than 5 gigawatts running over 2 million chips. OpenAI said again in September 2025 that Oracle, OpenAI and SoftBank were adding five new Stargate sites and that the program was ahead of schedule on a $500 billion, 10-gigawatt commitment by the end of 2025. (openai.com) An October update said the Midwest site would be in Wisconsin and developed by Oracle with Vantage. ### What do Oracle’s own numbers say about the shift? Oracle said on March 10, 2026 that third-quarter remaining performance obligations reached $553 billion, up 325% from a year earlier. (openai.com) The same release said cloud infrastructure revenue rose 84% year over year to $4.9 billion, while total cloud revenue reached $8.9 billion. Larry Ellison said in Oracle’s first-quarter fiscal 2026 results that the company expected to sign several additional multibillion-dollar customers and that remaining performance obligations were likely to exceed half a trillion dollars. (openai.com) Oracle also projected OCI revenue would grow to $18 billion in that fiscal year and then rise further over the following four years, with most of that forecast already booked in reported RPO, according to the company. (investor.oracle.com) ### What exactly is Oracle selling to AI customers? Oracle’s AI infrastructure page says OCI Supercluster can provide up to 131,072 GPUs for frontier-model training, inference and scientific computing. The company presents that offering as part of a distributed cloud model that can run demanding AI workloads “anywhere.” Oracle has also broadened the chip mix it offers. In June 2025, Oracle and AMD said Oracle would offer zettascale AI clusters using AMD Instinct MI355X GPUs, again at a scale of up to 131,072 GPUs. (investor.oracle.com) Oracle said the aim was to give customers more choice for training and inference workloads. ### Where is Oracle building this capacity? (oracle.com) Oracle said in January 2026 that it had AI infrastructure projects underway with OpenAI at two campuses in Texas and at sites in New Mexico, Michigan, Wisconsin and Wyoming. Josh Pitcock, an Oracle senior vice president, said the company was building infrastructure needed for AI-driven research and economic growth. (oracle.com) OpenAI’s September and October 2025 updates identified Oracle as a development partner on at least part of that buildout. The next public checkpoints are likely to come through Oracle earnings releases and future Stargate site updates from Oracle, OpenAI and SoftBank. (openai.com) (oracle.com)