The revolutionary computing platform adopted by top global robotics companies, enabling machine intelligence to truly integrate with the physical world for the first time.

At a critical juncture in the evolution of intelligent robotics, NVIDIA has once again introduced a groundbreaking innovation. On August 25, 2025, NVIDIA officially launched Jetson Thor, a next-generation computing platform specifically designed for physical AI and humanoid robots.

Hailed by NVIDIA founder Jensen Huang as the “ultimate supercomputer driving the era of physical AI and general-purpose robotics,” this product is redefining the level of intelligence and practical application capabilities of robots.

The “Most Powerful Brain” for Robots: Performance Leap Unleashes New Possibilities

Based on NVIDIA’s Blackwell GPU architecture, Jetson Thor delivers unprecedented computing performance. Compared to the previous generation Jetson Orin, Jetson Thor offers up to 7.5 times higher AI computing performance and up to 3.5 times better energy efficiency.

More notably, its CPU performance has improved by up to 3.1 times, meaning robots no longer require external CPUs, achieving complete integration of “brain and cerebellum” that provides a solid foundation for real-time computing.

This breakthrough design addresses one of the most significant challenges in robotics—simultaneously running multiple AI workflows—enabling robots to interact with humans and the physical world in real-time, intelligently.

Three Technological Innovations Unleashing Robotic Potential

1. Powerful Integrated CPU and Real-Time Computing Capability

Jetson Thor features a 14-core Arm Neoverse-V3AE CPU that can handle all computing tasks without an external CPU. This highly integrated design enables robots to achieve true real-time response, making rapid and precise decisions in dynamic environments.

For robot developers, this “brain and cerebellum” integrated design significantly simplifies system architecture, reduces power consumption and size, while improving system reliability and response speed.

2. HOLOSCAN Bridging Technology: Revolutionizing High-Speed Data Transmission

Jetson Thor incorporates NVIDIA’s innovative Holoscan sensor bridging technology, which can seamlessly convert various sensor data into network data. This technology provides up to 425Gb Ethernet bandwidth, supporting four 25 GbE network interfaces.

This means robots can synchronously process multimodal sensor data from lidar, cameras, microphones, and more, with latency below 10 milliseconds, meeting the real-time decision-making requirements of humanoid robots in complex environments.

3. JetPack 7.0: Quantum Leap in Development Experience

Jetson Thor comes with the latest JetPack 7.0 SDK, offering optimized data interfaces and development environment. This system is fully compatible with NVIDIA’s software stack from cloud to edge, including the Isaac platform for robot simulation and development, the Isaac GR00T humanoid robot foundation model, NVIDIA Metropolis for visual AI, and NVIDIA Holoscan for real-time sensor processing.

Developers can use the Isaac Sim simulation platform for digital twin training, then seamlessly deploy models to edge devices using JetPack 7 SDK, achieving an integrated “cloud-edge-device” development workflow.

Localized Large Model Deployment: Beyond Remote Control Toward Autonomy

Traditional robots often rely on remote control or cloud processing, limiting their performance in real-time applications. Jetson Thor fundamentally changes this with its 128GB large memory capacity and powerful local computing capability.

Now, robots can directly run mainstream language models including Llama, Gemini, Qwen, as well as vision-language-action models (VLA) and vision-language models (VLM) on edge devices. This capability truly liberates robots from dependence on remote controllers and stable network connections, enabling real-time processing at the edge.

In NVIDIA’s tests, Jetson Thor generated the first token in under 200 milliseconds and each subsequent token in under 50 milliseconds when running Llama 3B and Qwen 2.5 VL 3B models, equivalent to generating over 25 tokens per second. This response speed makes natural human-robot interaction a reality.

The Choice of Global Leading Enterprises and Research Institutions

NVIDIA Jetson Thor has already been adopted and recognized by numerous renowned global companies and research institutions:

• Agility Robotics, Boston Dynamics, and Amazon Robotics among other robotics leaders have integrated Jetson Thor into their next-generation robot systems
• Medical technology giant Medtronic is developing surgical assistant robots based on Jetson Thor for more precise medical operations
• Top research institutions including Stanford University, Carnegie Mellon University, and ETH Zurich are using Jetson Thor to advance research in perception, navigation, and planning models, attempting to break through real-time computing bottlenecks in fields such as medical triage and search and rescue
• Chinese robotics companies including Ubtech, Unitree Robotics, and Galaxy General have been among the first to adopt the Jetson Thor platform. Galaxy General founder and CEO Wang He stated: “After adopting Jetson Thor, our G1 Premium has achieved significant improvements in movement speed and action fluency”

Unlimited Application Scenarios: From Industrial Manufacturing to Healthcare

Jetson Thor’s powerful performance opens up new possibilities for robot applications across various industries:

In industrial automation, robots equipped with Jetson Thor can identify and grasp objects in real time while performing environmental reasoning through visual language models, avoiding obstacles and significantly improving production efficiency and safety.

In healthcare, United Imaging Healthcare has developed surgical assistant robots based on Jetson Thor that can real-time integrate 4K endoscope images with patient vital signs data, using GR00T models to predict surgical risks and assist doctors in minimally invasive operations. Tests show that this system reduces surgical decision-making time from minutes to seconds, improving operation precision by 30%.

In intelligent transportation, Wanji Technology‘s vehicle-road collaboration system utilizes Jetson Thor’s 10x I/O throughput to synchronously process data from 200 cameras and radars, achieving traffic flow prediction accuracy exceeding 95% and accident response time shortened to within 50 milliseconds.

Opening a New Era of Embodied Intelligence

The NVIDIA Jetson AGX Thor developer kit is now available, priced at $3,499. The Jetson T5000 production modules and Thor developer kit can be purchased through TWOWIN Technology.Twowin is NVIDIA NPN Partner, 11 years of industry experience.

NVIDIA Thor Developer Kit has arrived, please click here for consultation