NVIDIA has made it easier to customize its Nemotron 3 Nano AI model. With Prime Intellect Lab, setting up your own version takes about five minutes. This is a big step for developers who want to tweak open models quickly and locally.
The Nemotron 3 family includes three models: Nano, Super, and Ultra. They support workflows for baseline setup, training, and reevaluation. Prime Intellect Lab helps users apply reinforcement learning and deploy LoRA adapters, making fine-tuning more accessible.
Customization uses reinforcement learning with verifiable rewards (RLVR) inside a Python Math environment. NVIDIA also publishes open weights, data, and training recipes for Nemotron 3. This openness invites developers to experiment and improve AI models on their own terms.
NVIDIA Expands Physical AI with Omniverse Libraries
On July 20, 2026, NVIDIA announced the Agent Toolkit now includes Omniverse libraries. These libraries add physical AI capabilities to 3D tools developers already use. CEO Jensen Huang said, “The physical AI era will be built in simulation first.”
The new libraries cover RTX sensor simulation, GPU-accelerated physics with ovphysx, and CAD-to-SimReady skills. RTX sensor simulation generates camera, lidar, and radar outputs from 3D scenes. Ovphysx simulates realistic object behavior, while CAD-to-SimReady converts CAD data into simulation-ready assets.
Software makers like SideFX and PTC are integrating these tools. SideFX uses OpenUSD workflows, ovrtx, and ovphysx in Houdini. PTC connects design workflows to physical simulation with OpenUSD and ovrtx. SideFX’s CEO Kim Davidson said, “Procedural 3D creation is essential to building the complex, controllable worlds needed for simulation, robotics and industrial AI.”
A demo called “SimReady” built in Blender showcases how Omniverse libraries work together. Startups such as Palatial, Lightwheel, ForgeCAD, and Moonlake AI are already using the libraries to prepare assets and scenes.
New AI Hardware and Financing Models Drive Growth
This fall, RTX Spark systems will be available from ASUS, Dell Technologies, HP, Lenovo, Microsoft Surface, and MSI. Meanwhile, DGX Station systems can be ordered from ASUS, Dell, GIGABYTE, HP, MSI, Supermicro, and Exxact. These systems support the expanding AI workloads.
On the financing side, NVIDIA introduced a new model where it steps in if customers stop paying. This helps neoclouds like GMI Cloud secure financing they might not get otherwise. GMI Cloud committed $500 million to expand AI infrastructure using this model, making it one of the first in Asia to do so.
Other neoclouds like Firmus and Sharon AI also work with NVIDIA under this plan. NVIDIA’s partnership with CoreWeave and OpenAI combines investment and supply relationships, helping build a stronger AI compute ecosystem.
AI Infrastructure Management with Spectro Cloud
Spectro Cloud launched PaletteAI Inference Launchpad to help enterprises cut AI token costs by up to 70%. This system supports AMD-powered AI infrastructure. It includes AMD GPUs, GPU Operator, ROCm runtime, and AMD models. AMD’s Corporate VP Kumaran Siva explained, “Support for AMD-powered infrastructure in PaletteAI and PaletteAI Inference Launchpad helps customers accelerate production AI deployments across flexible, open AI stacks.”
PaletteAI also enables governance and local inference deployment. Spectro Cloud works with partners like NexusIgnite to extend this technology into regulated industries and sovereign environments. Spectro Cloud will demo PaletteAI at Booth 515A on July 22–23, 2026.
AI token consumption is forecasted to increase 24-fold by 2030, reaching 120 quadrillion tokens per month. This growth creates demand for scalable, cost-efficient AI infrastructure. NVIDIA and ecosystem partners are building tools, hardware, and financing to meet this challenge head-on.
Based on
- Start Customizing NVIDIA Nemotron 3 Nano with Prime Intellect Lab in Minutes — developer.nvidia.com
- NVIDIA Agent Toolkit Expands With New Omniverse Libraries, Putting AI Agents to Work Building Simulation-Ready Worlds | Markets Insider — markets.businessinsider.com
- Nvidia has a new way to sell more AI chips: help customers buy them | Business Insider Africa — africa.businessinsider.com
- I recreated OpenAI’s Codex Micro for a lot less – and mine is more customizable | ZDNET — zdnet.com
- Spectro Cloud Launches PaletteAI Inference Launchpad to Reduce Enterprise Token Costs by up to 70% | AFP.com — www-pp.afp.com
