NVIDIA AI Updates: July 2, 2026

1. NVIDIA Introduces Revenue-Sharing Model to Power AI Infrastructure Buildout

NVIDIA. NVIDIA unveiled a new business model that lets AI cloud companies procure its infrastructure through revenue-sharing partnerships, enabling large-scale multi-tenant AI factories for startups and enterprises. Initial partners include Sharon AI, deploying 40,000 GPUs, and Firmus, building capacity toward as many as 170,000 NVIDIA GPUs including a 360-megawatt facility in Indonesia. The approach aims to widen access to expensive compute while giving NVIDIA recurring, usage-based revenue as AI shifts from development to production inference at scale. Source

2. NVIDIA and Partners Expand American Manufacturing for AI Infrastructure

NVIDIA. NVIDIA said it and its manufacturing partners are investing in US-based production of AI infrastructure across 43 states, spanning semiconductor fabrication, electronics manufacturing, and AI data centers. Partners include TSMC in Arizona, Foxconn in Houston, and Wistron in the Dallas-Fort Worth area, alongside suppliers such as Corning, Lumentum, and Coherent. CEO Jensen Huang framed the buildout as a once-in-a-generation opportunity to reinvigorate American manufacturing while creating jobs in skilled trades and engineering. Source

3. NVIDIA Details Reinforcement Learning Techniques for AI Agents

NVIDIA. NVIDIA published a guide on using reinforcement learning with verifiable rewards and group relative policy optimization to customize language models and AI agents beyond prompting or supervised fine-tuning. Its NeMo RL ecosystem, including NeMo Gym and NeMo Data Designer, lets developers define task success criteria, generate training signals through verifiers, and update model weights so successful behavior becomes more likely. NVIDIA positions the approach as a way for teams building agents for workflows like security triage or customer support to improve reliability while keeping control of data and IP. Source

4. NVIDIA Uses Nsight Tools to Speed Up Omniverse NuRec Reconstruction Pipeline

NVIDIA. NVIDIA detailed how engineers used Nsight Systems and Nsight Compute to profile and optimize Omniverse NuRec, a neural reconstruction pipeline that builds high-fidelity 3D environments from camera and lidar data for simulation. Profiling exposed excessive kernel launches, unnecessary synchronization, and underutilized GPU resources, leading to a roughly 50x speedup in the interpolate function and kernel occupancy improvements from about 15% to 30-50%. The work shows how developer tools can uncover bottlenecks that reshape performance in demanding reconstruction workloads for autonomous vehicles and robotics. Source