The new Blackwell-based module delivers 7.5x the AI performance of its predecessor while fitting in a dramatically smaller footprint
Nvidia just pulled off the hardware equivalent of fitting a V8 engine into a go-kart. The company’s new Jetson AGX Thor module cuts the physical footprint of its robotics supercomputer roughly in half while cranking out 7.5 times the AI performance of its predecessor, the Jetson AGX Orin.
The module measures just 100 mm by 87 mm. It delivers up to 2,070 FP4 TFLOPS while sipping between 40 and 130 watts of power, which translates to 3.5 times the energy efficiency of the Orin generation.
What’s under the hood #
The Jetson AGX Thor is built on Nvidia’s Blackwell architecture, the same chip design powering the company’s latest data center GPUs.
For developers eager to get their hands on it, Nvidia is pricing the developer kit at $3,499 with expected availability around August 2025.
Humanoid robots and the physical AI push #
Nvidia announced new modules specifically designed for humanoid robots and edge AI applications, including the Isaac GR00T humanoid robot reference design. The Isaac GR00T platform is expected to become available in late 2026.
Nvidia is working with Boston Dynamics, Figure, and Unitree to accelerate deployment of advanced robotic systems across manufacturing, logistics, and other industries.
Why crypto investors should care about a robotics chip #
Nvidia’s stock has become a bellwether for the entire AI trade, and the AI narrative is deeply intertwined with crypto’s AI token sector. When Nvidia announces hardware breakthroughs, it validates the broader thesis that AI infrastructure is growing rapidly.
The edge computing angle is particularly relevant. Decentralized physical infrastructure networks, commonly called DePIN in crypto, are attempting to build distributed networks of compute nodes at the edge. A $3,499 developer kit with 2,070 TFLOPS of compute is the kind of hardware that DePIN node operators dream about.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our