Nvidia makes Jetson AGX Thor generally available. The module delivers up to 7.5 times more AI computing power and 3.5 times better energy efficiency than its predecessor Jetson Orin.
Nvidia launches Jetson Thor. It’s a brain based on the Blackwell architecture for AI-driven robots. Jetson Thor aims to improve how mechanical robots interact with humans and their environment. The chip contains a Blackwell GPU, 128 GB memory, and delivers up to 2,070 FP4 teraflops of AI compute with 130 watts power consumption. This makes the hardware suitable for driving multiple generative AI models in real-time, such as vision-language-action models and large language models.
Jetson Thor succeeds Jetson Orin. According to Nvidia’s own figures, the new chip is 7.5 times more powerful and 3.5 times more energy-efficient. Nvidia has been working on Jetson for robots since 2018.
Ecosystem
The module runs on the Nvidia Jetson software platform. It’s compatible with popular AI frameworks and supports applications from cloud to edge. Jetson Thor is intended for use in manufacturing, logistics, healthcare, agriculture, and retail, among others.
Jetson Thor aims to provide a solution for what Nvidia calls one of the biggest challenges in robotics: combining multiple AI processes on a single platform for real-time decision making. This allows robots to respond faster in unpredictable situations.
Nvidia, for its part, is trying to control AI development as much as possible. By focusing on hardware, software, and an ecosystem, the company has made itself de facto indispensable in numerous AI implementations, from the data center to robots.
Enthusiasm
According to Nvidia, more than 2 million developers worldwide use the Jetson environment, and the ecosystem consists of more than 150 partners. Jetson Thor builds on the success of Jetson Orin, which is now used by 7,000 customers.
The Jetson AGX Thor developer kit is available starting at $3,499. Production modules, such as the Jetson T5000, are available through international distribution partners and embedded suppliers.