NVIDIA has officially released Jetson Thor, its most powerful robotics computer to date, built to revolutionize humanoid robots and physical AI systems.
TLDR:
- Jetson Thor delivers 7.5 times more AI compute power than its predecessor
- The platform enables real-time decision making on humanoid and industrial robots
- Major adopters include Amazon, Meta, Caterpillar, OpenAI, and Agility Robotics
- NVIDIA stock rose 1.02% as investors reacted to its growing dominance in robotics
What Happened?
NVIDIA announced the general availability of Jetson Thor, a powerful AI computer designed for robotics and autonomous machines. The system marks a leap in on-device processing with up to 2,070 FP4 teraflops of compute and advanced edge capabilities, unlocking new potential for real-time reasoning in robots.
The NVIDIA Jetson Thor is here. 🎉
— NVIDIA Robotics (@NVIDIARobotics) August 25, 2025
This powerful new robotics computer is designed to power the next generation of general and #HumanoidRobots in manufacturing, logistics, construction, healthcare, and beyond.
It’s a massive leap forward for physical AI.
Early adopters… pic.twitter.com/zBwDZkeRCM
Jetson Thor Unlocks Smarter Robots
Jetson Thor is built with NVIDIA’s Blackwell GPU and includes a massive 128GB of memory, delivering more than seven times the AI compute power and twice the memory of its predecessor, Jetson Orin. Designed to handle multiple AI models at once, including large language and vision-language-action models, Jetson Thor enables robots to perceive, think, and act faster and more efficiently.
NVIDIA says Jetson Thor operates within a 130-watt power envelope, making it 3.5 times more energy efficient while offering 3.1 times greater CPU performance. This allows for richer sensor data processing and faster visual reasoning on the edge without relying on cloud infrastructure.
Wide Industry Adoption
Adoption is already gaining speed across multiple sectors:
- Agility Robotics plans to use Jetson Thor in the sixth generation of its Digit robot for warehouse logistics
- Boston Dynamics is integrating it into the humanoid robot Atlas
- Amazon Robotics, Caterpillar, Meta, John Deere, OpenAI, and Medtronic are actively testing or deploying the platform
NVIDIA’s software stack, including Isaac for robotics, Metropolis for vision AI, and Holoscan for sensor processing, is fully compatible with Jetson Thor, giving developers tools to build and deploy AI-driven robotic systems quickly.
From Research Labs to the Real World
Leading academic institutions such as Stanford, Carnegie Mellon, and University of Zurich are integrating Jetson Thor in cutting-edge robotics research. For example, Carnegie Mellon’s robotics team will upgrade their medical triage robots with Jetson Thor to enhance real-time navigation and decision-making.
Jetson Thor also supports generative reasoning models like Cosmos Reason, DeepSeek, Llama, Gemini, Qwen, and robotics-specific models like Isaac GR00T N1.5, enabling robots to adapt and learn in dynamic environments.
Developer Kits and Market Availability
The Jetson AGX Thor developer kit, which includes a T5000 module, power supply, and cooling system, is now available starting at $3,499. The T5000 modules alone are priced from $2,999 for bulk purchases.
NVIDIA also opened preorders for Drive AGX Thor, a version tailored for autonomous vehicles, with shipments expected in September.
What TechKV Thinks?
I think this is one of NVIDIA’s boldest and smartest moves yet. Jetson Thor isn’t just a chip; it’s the missing piece that makes advanced robotics truly autonomous. We’ve seen robots do impressive things, but they’ve always been limited by latency and cloud dependence. With Jetson Thor, that ceiling just got smashed. It puts server-grade AI right into robots. And the fact that it’s already being adopted by both startups and giants like Amazon and Caterpillar shows this is more than hype. It’s happening.
