Oracle has made the general availability of its Nvidia Vera Rubin NVL72 systems and standalone Vera CPUs on its cloud infrastructure effective Oct 27. The cloud provider outlined an initial supply of 150,000 Vera CPUs for the current quarter, enabling customers to scale compute and GPU resources independently for the first time. OpenAI is bringing these processors into production to run code and tools within its agentic sandboxes, citing tests that showed 2.2x memory bandwidth per thread and 51% higher Python throughput compared to leading x86 CPUs.
The launch is part of a broader expansion of Oracle Cloud Infrastructure (OCI) to include a variety of specialized hardware, including a plan to scale Vera Rubin superclusters to multi-gigawatt capacity and the introduction of HGX Rubin NVL8 systems featuring Intel Xeon 6 CPUs. Oracle is also deepening its chip diversification by adding inference systems from Cerebras and SambaNova, and it will become the first hyperscaler outside Google to host custom TPU 8t and 8i chips starting in 2027.