According to Nvidia's announcement, the expansion targets training and serving large models as well as the tasks AI agents perform between model calls, such as running tools, testing code and retrieving data. Customers will be able to scale CPU and GPU resources independently. Oracle also offers Oracle Acceleron, which provides hardware-based isolation for workloads running AI-generated code.
The planned HGX Rubin NVL8 server includes eight Rubin GPUs connected through NVLink, two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps of RDMA networking, in an x86-based configuration.
The announcement says OpenAI trained and served GPT-6 Astra on an OCI Supercluster in Abilene with more than 100,000 Blackwell GPUs, and that OpenAI is bringing Vera CPUs into production on OCI for its agent sandbox workloads. The announcement does not give the exact power of the expansion, a timeline or pricing.