Microsoft is using the Oracle Cloud to provide sufficient GPU computing power for OpenAI. Oracle is said to be building a large data center packed with accelerators.

To meet OpenAI’s immense hardware requirements, Microsoft is extending the Azure AI platform to Oracle Cloud Infrastructure (OCI). This is a notable move and shows how hungry OpenAI is for GPU computing power. After all, Microsoft doesn’t have much to gain from this move unless OCI has enough internal GPU accelerators while Microsoft looks for more.

Oracle is said to be building a large data center full of Nvidia chips. At least that’s what CTO Larry Ellison said. Half of this data center would be used by Microsoft.

Connect

OCI and Azure have had a privileged relationship for some time. For example, Oracle Interconnect for Azure has been available for several years, allowing customers to establish superfast connections between both data centers with sub-2ms latency.

However, the collaboration now announced is of a different nature. Microsoft will expand its services to OCI to meet the needs of OpenAI, in which it has invested so much. This shows how scarce Nvidia’s most powerful NPUs are. You can be sure that Microsoft would prefer to have the capacity in-house if it could buy enough GPUs.