Oracle unveils OCI's August AI batch: Nemotron on day zero, H100 multi-node serving
OCI Enterprise AI picked up NVIDIA's Nemotron 3.5 Lightning on day zero; H100 multi-node serving, background mode for the Responses API and natural-language database access round out the batch.
Oracle has announced the August batch of enterprise AI updates for its OCI cloud. The first headline sits on the model side: OCI Enterprise AI became one of the first clouds to offer NVIDIA's Nemotron 3.5 Lightning, the open model designed for always-on agents, on day zero (as of August 11).
On infrastructure there are two concrete steps. H100 multi-node serving is now open for imported models: large models can be served across multiple H100 GPU nodes, with Oracle absorbing the burden of managing the distributed setup. The on-demand model catalog has expanded as well, and Cohere and Meta models now come with additional hardware shape choices.
Two more items matter to developers: the OCI Responses API gained background mode, so long-running reasoning, document processing and agentic tasks can execute asynchronously without holding a connection open. Hosted application endpoints now also authenticate through OCI's Identity and Access Management, letting companies apply their existing policy framework to AI endpoints.
The quiet star of the batch is on the data side: natural-language access to data has arrived in the Oracle Database Console and OCI's SQL assistant. The translation layer everyone runs in their head, thinking in plain language and writing SQL, now lives inside the database console itself.
There is a sovereignty leg to the batch as well: OCI Enterprise AI is now available in dedicated cloud regions, letting organizations adopt production AI while keeping sensitive data and workloads inside their own security, governance and sovereignty boundaries. It is the enterprise-cloud counterpart of the in-country inferencing wave we saw in India this week.
