NVIDIA Introduces New Jetson Thor Models for Physical AI
NVIDIA introduced the Jetson Thor T3000 and T2000 models for physical artificial intelligence projects. Revolutionary features for humanoid robots and edge artificial intelligence are in our news.
NVIDIA introduced the new Jetson Thor T3000 and T2000 solutions, which it developed to strengthen the physical artificial intelligence ecosystem. With this new hardware, the company aims to provide a scalable infrastructure for humanoid robots, industrial robotic systems and edge artificial intelligence applications. These platforms, announced at the event in Japan, promise high computing performance and power efficiency, especially by being powered by the Blackwell GPU architecture. With this move, NVIDIA aims to accelerate the spread of robotic technologies around the world and provide technological support to the physical artificial intelligence (Physical AI) processes of industry giants such as Boston Dynamics and Amazon Robotics.
- The Jetson Thor T3000 model offers 865 TFLOPs of AI computing power for humanoid robots.
- The T2000 version is an entry-level solution for edge AI applications with 400 TFLOPs performance.
- NVIDIA significantly reduces the memory usage of devices thanks to software optimizations.
- Both new modules are planned to be released in the first quarter of 2027.
Jetson Thor T3000 Meets High Performance Needs
NVIDIA Jetson Thor T3000 is positioned as a more compact and optimized version of the flagship T5000 series. Equipped with Blackwell GPU architecture, this system has 865 TFLOPs FP4 computing capacity. Operating with a power consumption of approximately 70 Watts, the T3000 provides the high memory bandwidth required for large language models and multi-mode artificial intelligence tasks with 32 GB LPDDR5X memory.
The new platform takes the capacity of robotic systems to make sense of the complex world to the next level.
Jetson Thor T2000 Increases Productivity at the Edge
Jetson Thor T2000, designed for lighter workloads and portable autonomous robots, stands out with its 400 TFLOPs artificial intelligence performance. This model, which has a memory capacity of 16 GB, has a very low power consumption of 40 Watts. This efficiency provides a great advantage, especially in portable robotic applications where battery life is critical.
NVIDIA Saves Memory with Software Optimizations
In addition to hardware innovations, NVIDIA increases the efficiency of existing Jetson devices with software improvements. New Jetson agent skills allow optimization of the software stack, saving up to 50% in memory usage.
Software optimizations allow businesses to get the same performance with lower-cost hardware.
Leading companies such as UBTech and Agile Robots have achieved memory savings of up to 15 GB using these optimizations. This makes it possible for companies to switch to more cost-effective solutions instead of higher segment hardware. While the emulation mode of Jetson Thor T3000 will be available to developers via JetPack 7.2.1 this month, general shipment of the modules will begin in the first quarter of 2027.
Do you think this new hardware attack by NVIDIA will create a big breakthrough in the world of robotics? You can share your ideas and expectations about the future of physical artificial intelligence with us in the comments section.
