Microsoft will once again turn to Oracle to help OpenAI expand its computing power. According to The Register, the company leases AI power from Oracle Cloud Infrastructure (OCI). Microsoft has already invested heavily in OpenAI and, according to some sources, owns almost half of the company, but it seems that Azure resources are not enough for the AI startup to grow freely.
Oracle’s Data Center Expansion
Oracle’s latest quarterly report stated that the company would soon build a “very, very large data center,” about half of which would be used by Microsoft. It is expected to employ many NVIDIA accelerators with LSS, intended primarily for training AI models. Oracle’s second-generation AI infrastructure will be significantly more productive than the first, and there will be great demand for it, according to Oracle.
The OCI Supercluster will be able to combine up to 64 thousand NVIDIA Blackwell GB200 super accelerators via an RDMA interconnect with ultra-low latency and will receive high-performance storage. In addition to Microsoft, the supercluster will be used by companies like Adept, Modal, MosaicML, NVIDIA, Reka, Suno, Together AI, Twelve Labs, and xAI. Elon Musk has already announced his readiness to spend $10 billion on Oracle cloud servers. At the same time, Oracle offers its own OCI Generative AI service, which is a competitor to Azure OpenAI.
Growing Partnership and Future Developments
The first rumors about the mutual use of cloud resources between Microsoft and Oracle appeared a year ago. Subsequently, Microsoft actually leased NVIDIA accelerators from Oracle. Additionally, the companies have become much closer—Oracle has placed its Exadata complexes in Azure. A similar solution will soon appear in the Google Cloud Platform with the latest Cross-Cloud Interconnect service, which will be available in 11 OCI cloud regions. The Oracle Database@Google Cloud project is expected to start by the end of 2024.
In the last quarter, Oracle’s fastest-growing revenue was from its cloud licensing support services. The company announces the largest contracts in history amid demand for training AI models and record levels of sales of OCI services. At the same time, customers are moving from one-time purchases to multi-year cloud services, which is what Oracle sought, adds NIXsolutions. The company’s RPOs have already reached $98 billion.
In the near future, Oracle will increase the performance of its AI infrastructure to support the latest models and build more large data centers, with the capacity of some approaching 1 GW. Microsoft and OpenAI were rumored to be exploring a $100 billion 5GW campus in March 2024. Microsoft is also believed to be one of CoreWeave’s largest customers. We’ll keep you updated as Microsoft continues to rapidly increase the capacity of its data centers, commissioning five AI supercomputers every month.