NVIDIA AI Updates: June 27, 2026

1. NVIDIA AI-Q Blueprint Ships for Production Agents on Oracle Cloud

NVIDIA detailed how to deploy its AI-Q Blueprint, an open source reference for long-horizon AI agents built on LangChain Deep Agents and the NVIDIA NeMo Agent Toolkit, on Oracle Cloud Infrastructure. The walkthrough uses Terraform to provision infrastructure and Helm to install the Kubernetes workloads, producing a production-ready multi-agent endpoint with automated teardown. The guide matters because it lowers the operational barrier to running persistent, multi-step agent systems on managed cloud infrastructure. Source

2. NVIDIA Model Optimizer Quantizes Nemotron 3 Ultra to NVFP4

NVIDIA described creating an NVFP4 checkpoint for its 550B-parameter Nemotron 3 Ultra model using the NVIDIA Model Optimizer. Post-training quantization compresses the model weights from 1,121 GB down to roughly 352 GB while maintaining accuracy, enabling faster inference and a smaller memory footprint on NVIDIA hardware. The reduction makes large frontier models more practical to serve, since lower-precision formats cut the GPU memory and bandwidth needed per deployment. Source

3. KRAFTON Builds PUBG Ally Co-Playable Character on NVIDIA ACE

NVIDIA and KRAFTON detailed PUBG Ally, an AI-powered teammate character that combines automatic speech recognition, a small language model, and text-to-speech to interact with players during matches. Powered by NVIDIA ACE, the character runs locally on players’ devices and pairs voice understanding with real-time game awareness to deliver contextual assistance without cloud latency. The example shows ACE moving conversational, on-device game characters from demo to shipping titles. Source