Jetson AGX Thor is now official: this is NVIDIA's kit to give real autonomy to industrial, medical, and humanoid robots.

Last update: 27/08/2025

  • Jetson AGX Thor arrives with T5000 and T4000 modules and a development kit focused on advanced robotics.
  • Up to 2.070 TFLOPS FP4, 7,5x increase in AI and 3,5x improvement in efficiency over Orin.
  • Extensive connectivity: PCIe Gen5, USB-C/USB-A, HDMI/DP, 5GbE and QSFP28 4x25GbE in the dev kit.
  • Complete ecosystem with NVIDIA Isaac, Metropolis, and Holoscan, and adoption by leaders like Amazon Robotics and Boston Dynamics.

NVIDIA Jetson AGX Thor platform for robotics and AI

NVIDIA launches Jetson AGX Thor developer kit together with its production modules, a AI edge computing platform designed for manufacturing robots, Logistics, transportation, healthcare, agriculture and retailThe objective is clear: Bringing generative AI and multimodal reasoning to robots so that they can interact with people and environments in real time.

NVIDIA management emphasizes that Those who create robotic systems rely on their platforms to combine high performance and energy efficiency. and thus run multiple AI models at the edge

Power and efficiency for physical AI at the edge

Jetson AGX Thor Module AI Performance

The leap compared to the previous generation is overwhelming: up to 7,5 times more AI computing power and an improvement of 3,5 times more energy efficient versus Jetson AGX Orin. The platform is ready for multi-AI workflows, included vision-language-action models latest batch as Isaac GR00T N1.5 and great language models.

The Jetson T5000 integrates a GPU NVIDIA Blackwell 2.560 CUDA Cores and 96 Tensor Cores, CPU of 14 ARM cores at 2,6 GHz, and 128GB LPDDR5X with up 273 GB / s of bandwidth. In figures, it reaches 2.070 TFLOPS in FP4 with a reference consumption of 130 W, a solid foundation for robots that require instant perception and decision-making.

Exclusive content - Click Here  OpenAI strengthens Sora 2 after criticism from Bryan Cranston: new barriers against deepfakes

For those who prioritize efficiency, the Jetson T4000 offers up 1.200 TFLOPS with Blackwell GPU from 1.525 CUDA Cores and 64 Tensor Cores, CPU of 12 ARM cores at 2,6 GHz, and 64GB LPDDR5X keeping the 273 GB/s bandwidth, with an energy profile around 70 W.

Designed for physical AI, both modules They allow you to simultaneously process inputs from cameras, LiDAR, and other sensors, run generative and control models, and respond with low latency., something key in humanoids, drones, industrial arms or surgical assistants.

Jetson AGX Thor Development Kit and High-Level Connectivity

Jetson AGX Thor Development Kit and Motherboard

El Jetson AGX Thor Developer Kit integrates the module T5000 and a motherboard with abundant connectivity to accelerate testing, sensor integration, and production deployment. The package includes active dissipation and feeding so that the deployment is immediate.

  • Storage/Expansion: two slots M.2 (PCIe Gen5 x4 and Gen5 x1) for NVMe SSDs or other peripherals.
  • Ports: 2x USB-A 3.1, 2x USB-C 3.1, HDMI 2.0b y Display Port 1.4a.
  • Red: RJ45 Ethernet 5 GbE y QSFP28 with up to 4×25 GbE.
  • Additional connectivity: M.2 for Wi-Fi, pin headers, audio and camera interface ready for prototyping.
Exclusive content - Click Here  What is a VGA port and what is it used for?

All of this is integrated into a compact form factor of 243,19 × 112,40 × 56,88 mm and a TDP range of 40 to 130 W, suitable for test benches, mobile robots and laboratory validation stations.

Ecosystem, models and adoption in the industry

NVIDIA Isaac, Metropolis, and Holoscan Ecosystem

Jetson Thor relies on the software stack NVIDIA Jetson for physical and robotic AI, compatible with major generative AI frameworks and with native integration with NVIDIA Isaac (simulation and robotics), Metropolis (computer vision) and Holoscan (real-time sensor processing).

The Jetson ecosystem has grown over the years to bring together more than 2 million developers and a fabric of 150+ partners in hardware, software and sensors. With Jetson Orin already operate more than 7.000 clients Across multiple sectors; Thor takes the next step to enable more complex systems, including humanoids and surgical robotics.

Companies like Agility Robotics, Amazon Robotics, Boston Dynamics, Caterpillar, Figure, Hexagon, Medtronic and Meta already work with the platform, and others like John Deere and OpenAI evaluate their fit in physical AI projects. The shared vision is that advanced models run on the robot itself, reducing latencies and cloud dependency.

From the product perspective it is highlighted that deploy generative models at scale at the edge allows robots to perceive, reason and act with greater autonomy, and that the The combination of computing power and efficiency is key to operating safely in dynamic logistics and industrial environments.

Exclusive content - Click Here  Google Images: New features in Photos, Gemini, and the leap to Nano Banana 2

Availability and prices

El Jetson AGX Thor Developer Kit It can now be purchased from $3.499Production modules are available through the distributor network: The T5000 starts at $2.999 on orders of 1.000 units or more and the T4000 from $1.999, depending on configurations.

For automotive, NVIDIA has announced a variant geared towards autonomous driving, with DRIVE/Thor open reservations for developers which will incorporate the same modules T5000 and T4000. Manufacturers can supplement with motherboards and production systems offered by integration partners.

With global availability of hardware and a consolidated software stack, Tests in the kit are quickly moved to production, shortening development cycles and facilitating scalability from prototypes to real deployments.

Combining AI power, industrial connectivity and a mature ecosystem places Jetson AGX Thor as a solid choice for those who need robots capable of perceiving, understanding, and acting in real time., from assembly lines and warehouses to operating rooms and farms.

Paralyzed man controls robotic arm-0
Related article:
Paralyzed man controls robotic arm with his mind thanks to new interface