NVIDIA Makes First Major Move in ‘Physical AI’ With Jetson Thor, Giving Humanoid Robots Their Own Brain
NVIDIA has officially unveiled Jetson Thor, its next-generation Jetson platform built on the Blackwell architecture, ushering in what CEO Jensen Huang calls the era of “Physical AI.” The company envisions humanoid robots equipped with advanced AI brains, capable of autonomous reasoning and decision-making, marking one of the boldest steps yet toward practical, real-world robotics.
Jetson Thor: A Major Leap From Jetson Orin
Detailed in NVIDIA’s official blog, Jetson Thor introduces the Jetson T5000 module, which packs enormous compute power:
GPU: 2560-core Blackwell-based GPU with 96 fifth-gen Tensor Cores
CPU: 14-core Arm Neoverse V3AE 64-bit processor
Memory: 128 GB LPDDR5X, 256-bit, delivering 273 GB/s bandwidth
Storage: 1 TB NVMe (M.2 Key M slot)
Networking: 1x 5GbE RJ45, 1x QSFP28 (100 GbE)
I/O: HDMI, DisplayPort, 2x USB-A 3.2, 2x USB-C 3.1, Gigabit Ethernet, dual CAN headers
Power Consumption: 40W–130W range
Performance improvements over Jetson Orin are dramatic:
7.5x more AI compute
3.1x faster CPU performance
2x more memory capacity
In raw terms, the Jetson Thor can deliver up to 2,070 FP4 teraflops of AI compute, all within 130 watts of power consumption, making it a breakthrough for real-time reasoning and generative AI deployment inside humanoid robots.
Industry Adoption and Pricing
Two of the biggest names in robotics, Agility Robotics and Boston Dynamics, have already begun deploying Jetson Thor into their next-gen platforms, potentially accelerating the arrival of more capable humanoids. Meanwhile, NVIDIA’s longtime partner Foxconn is preparing to bring humanoid robots to market by the end of this year, showing how quickly “Physical AI” could move from concept to reality.
However, there is a price to match the power. NVIDIA has set the Jetson AGX Thor developer kit at $3,499, positioning it firmly as a premium platform for serious robotics research and commercial deployment.
In a symbolic gesture, Jensen Huang personally signed several of the first Jetson Thor units delivered to NVIDIA customers, underscoring the importance he places on this new chapter in AI.
Physical AI: The Next Era
NVIDIA sees Physical AI as the next frontier, where robots infused with generative AI capabilities will move beyond automation and into autonomous reasoning. By giving humanoids a powerful brain, NVIDIA hopes to unlock robotics applications across manufacturing, logistics, healthcare, and domestic assistance.
While adoption will depend on cost and real-world integration, Jetson Thor positions NVIDIA as the clear frontrunner in providing the AI compute backbone for humanoid robotics.
The question now: Will “Physical AI” truly become the next big revolution after AI training and inference, or will high costs slow its widespread adoption?
