Enterprise and physical AI are about to get a significant boost. According to Capgemini, AI agents could generate up to $450 billion in additional revenue and cost savings by 2028. To reach this potential, leading developers and companies are turning to more powerful reasoning models that can provide systems with greater autonomy, contextual understanding, and decision-making capabilities.
At SIGGRAPH, NVIDIA announced the expansion of its Nemotron and Cosmos Reason model families, designed to enhance both enterprise AI agents and humanoid robots, as well as physical AI systems. Companies already working with these technologies include CrowdStrike, Uber, Magna, NetApp, and Zoom.
Nemotron: Maximum Precision and Efficiency for Enterprise AI
The NVIDIA Nemotron family consists of advanced reasoning models that combine hybrid architectures, NVIDIA-curated datasets, and cutting-edge AI techniques to deliver market-leading accuracy and efficiency.
Notable updates include:
- Nemotron Nano 2: delivers up to six times more token generation than other models of similar size.
- Llama Nemotron Super 1.5: the most precise reasoning in its class, now in NVFP4 (4-bit floating-point) format, with six times the performance on NVIDIA B200 GPUs compared to H100.
- 60% reduction in reasoning costs: enabled by configurable thought budgets and compact quantized models.
These models enable AI agents to explore more options, speed up research processes, and deliver more accurate results without increasing resource consumption.
NVIDIA has also introduced its first open dataset for training vision-language models (VLMs): Llama Nemotron VLM Dataset v1, containing 3 million samples for OCR, visual question-answering, and subtitle generation.
Cosmos Reason: Advanced Reasoning for Physical AI and Robotics
Cosmos Reason is a 7-billion-parameter vision and reasoning (VLM) language model designed to give robots and AI agents a deep understanding of the physical world.
Its ability to incorporate prior knowledge, physics, and common sense makes it ideal for applications like:
- Annotating training data for robotics and autonomous driving.
- Planning and decision-making with vision-language-action (VLA) models.
- Video analysis in industrial and urban environments to detect anomalies and optimize operations.
Companies such as Uber, Magna, and Ambient.ai are already using Cosmos Reason for vehicle behavior analysis, urban delivery planning, and safety supervision.
Use Cases in Business and Public Sector
- Zoom: integrated Nemotron so Zoom AI Companion can make decisions and manage complex tasks in meetings, chats, and documents.
- CrowdStrike: utilizes Nemotron for their Charlotte AI agents to generate queries within the Falcon platform.
- EY: adopting Nemotron Nano 2 for tax, risk management, and finance use cases.
- Uber: post-training Cosmos Reason to interpret visual data, enhancing autonomous vehicle safety.
- VAST: deploying Cosmos Reason for real-time urban intelligence, analyzing massive video streams and activating automatic responses.
New RTX PRO Blackwell GPUs for Professional AI Workflows
NVIDIA also announced two new GPUs based on the Blackwell architecture:
- RTX PRO 4000 Blackwell SFF Edition: compact form factor, 70W, up to 2.5x more AI performance and 1.7x better ray tracing compared to previous generation.
- RTX PRO 2000 Blackwell: optimized for design and 3D modeling workflows, offering 1.6x faster rendering and improved image and text generation for faster iteration and prototyping.
These cards target sectors like engineering, architecture, media, entertainment, healthcare, and public administration, with use cases spanning from flood simulation in Denver to urban design with AI in London.
Software Ecosystem and Availability
Nemotron and Cosmos Reason models will be available as NVIDIA NIM microservices, ensuring secure deployment on NVIDIA-accelerated infrastructure, and through platforms like Amazon Bedrock, Azure AI Foundry, Oracle Data Science Platform, and Google Vertex AI.
The RTX PRO 2000 and 4000 Blackwell GPUs will be released this year via manufacturers such as Dell, HP, and Lenovo.
FAQs
What sets Nemotron apart from other reasoning models?
Its hybrid architecture and token generation optimization make it faster, more accurate, and cost-efficient.What does Cosmos Reason bring to robotics?
It introduces physical and spatial reasoning, allowing robots and AI agents to understand, plan, and act with common sense.Which sectors can benefit from these solutions?
From cybersecurity and banking to logistics, automotive, and advanced manufacturing.Are these models compatible with edge deployments?
Yes, Cosmos Reason can run on compact devices like the RTX PRO 4000 SFF, bringing advanced physical AI to the network’s edge.
via: blogs.nvidia.com