Agentic AI Moves to the Edge
NVIDIA has announced a major update to its Jetson platform, designed to bring agentic AI capabilities to physical devices like robots, industrial inspection systems, and automation equipment. The release includes JetPack 7.2 and support for the NemoClaw agentic AI framework on Jetson hardware. This update allows developers to run sophisticated AI agents directly on edge devices rather than relying solely on cloud servers.
Jetson is a multi-generation platform powering edge AI in robotics, autonomous systems, and medical devices. With these additions, NVIDIA aims to provide a production grade stack that can handle real time decision making and automation tasks in physical environments. The announcement was made at COMPUTEX, where the company also highlighted the GTC Taipei Build a Claw event that demonstrates how agentic AI can function as a personalized always on assistant on Jetson hardware.
Key Features and Performance Gains
JetPack 7.2 introduces several important upgrades. It includes Yocto based operating system support, giving industrial customers a more customizable and memory efficient Linux foundation. The update also brings CUDA 13 to Jetson Orin modules and adds Multi Instance GPU (MIG) support to the Jetson Thor platform. MIG combined with a real time kernel allows developers to reserve dedicated GPU resources for time sensitive tasks like robot perception systems.
The Jetson AGX Orin 32GB module receives a performance boost to 241 TOPS of AI compute, a 20% increase over its original specification. A new middle layer of agent skills is included to automate developer tasks such as Linux customization and system building. This layered approach lets developers deploy physical AI agents in production environments more quickly while reducing total cost of ownership.
Source: NVIDIA Blog
