Oracle and NVIDIA unveil Zettascale10 and in-database AI

Oracle and NVIDIA are expanding their alliance to make enterprise AI faster to deploy and easier to scale across real business data. Announced at Oracle AI World, the centerpiece is OCI Zettascale10, a new NVIDIA-accelerated cluster designed for training and inference at extreme scale. Oracle says the platform delivers up to 16 zettaflops of peak AI performance and uses NVIDIA Spectrum-X Ethernet so GPUs spend less time waiting on data, allowing organizations to scale to millions of processors.

Oracle highlights “multi-gigawatt” capacity and native access to NVIDIA AI Enterprise on OCI, giving customers a ready toolset alongside more than 200 cloud services. The software strategy puts AI next to the data rather than moving data to the model. Oracle AI Database 26ai integrates AI directly into the database layer so teams can build agentic AI workflows that combine private enterprise information with outside sources without moving sensitive data.

A Unified Hybrid Vector Search spans relational, JSON, and spatial data to give AI consistent context, while security adds NIST-approved quantum-resistant encryption for data in transit and at rest. Designed to work with NVIDIA’s ecosystem, the database now connects to NVIDIA NeMo Retriever microservices to simplify retrieval-augmented generation, improving accuracy by grounding responses in company documents.

Oracle’s Private AI Services Container will offload vector-embedding creation to NVIDIA GPUs via the cuVS library to speed data preparation for AI search. Beyond the database, the Oracle AI Data Platform adds a built-in NVIDIA GPU option and the NVIDIA RAPIDS Accelerator for Apache Spark so data teams can speed pipelines and machine learning with little or no code changes.

All capabilities are organized in Oracle AI Hub, which lets users deploy NVIDIA NIM microservices through a simple, no-code interface. NVIDIA AI Enterprise is now available natively in the OCI Console, letting developers spin up GPU instances and enable NVIDIA software in a few clicks. Together, Oracle and NVIDIA aim to reduce friction across hardware, data, and tools, signaling a push toward practical, secure, and scalable enterprise AI.