I. Product Features:
1.1 Performance and Hardware Upgrades
The NVIDIA® Jetson AGX Thor™ Developer Kit powered by the NVIDIA Blackwell GPU and 128 GB of memory, delivering up to 2070 FP4 TFLOPS of AI compute to effortlessly run the latest generative AI models—all within a 130 W power envelope. Compared to NVIDIA Jetson AGX Orin™, it provides up to 7.5x higher AI compute and 3.5x better energy efficiency.
Jetson AGX Thor helps you accelerate low-latency, real-time applications with the new Blackwell Multi-Instance GPU (MIG) technology and a robust 14-core Arm® Neoverse®-V3AE CPU. It also includes a suite of accelerators, including a third-generation Programmable Vision Accelerator (PVA), dual encoders and decoders, an optical flow accelerator, and more. For high-speed sensor fusion, the developer kit offers extensive I/O options, including a QSFP slot with 4x25GbE, a wired Multi-GbE RJ45 connector, multiple USB ports, and additional connectivity interfaces. It’s also designed for seamless integration with existing humanoid robot platforms, allowing for easy tethering to jumpstart prototyping.
1.2 Real-Time Interaction and Multimodal Data Processing
Jetson Thor is specifically designed for generative inference models, supporting next-generation physical AI agents. These agents are powered by large transformer models, visual language models, and visual language action models, enabling real-time operation at the edge while minimizing reliance on cloud computing. This allows robots to go beyond perception and achieve reasoning and physical intelligent behavior.
With NVIDIA Holoscan sensor bridging technology and four 25 GbE high-speed network interfaces, Jetson AGX Thor can synchronously process data from various sensors such as LiDAR, cameras, and microphones, with a latency of less than 10 milliseconds.This means robots can synchronously process multimodal sensor data from lidar, cameras, microphones, and more, with latency below 10 milliseconds, meeting the real-time decision-making requirements of humanoid robots in complex environments.
1.3 Software Ecosystem and Full-Stack Toolchain
he Jetson Thor module also supports the complete NVIDIA AI software stack, accelerating nearly all physical AI workflows across platforms including NVIDIA Isaac for robotics, NVIDIA Metropolis for video analytics AI agents, and NVIDIA Holoscan for sensor processing.
With these software tools, developers can easily build and deploy various applications, such as: visual AI agents that analyze real-time camera streams to monitor worker safety, humanoid robots that perform operational tasks in unstructured environments, and intelligent operating room systems that use multi-camera stream data to guide surgeons.
Optimized through the Jetson software stack, Jetson Thor meets the low-latency and high-performance demands of real-time applications while supporting all mainstream generative AI frameworks and AI inference models, delivering significant real-time performance advantages. These models include general-purpose models like Cosmos Reason, DeepSeek, Llama, Gemini, and Qwen, as well as robotics-specific models like Isaac GR00T N1.5, enabling developers to effortlessly experiment with and run inference locally.
1.4 Powerful Integrated CPU and Real-Time Computing Capability
Jetson Thor features a 14-core Arm Neoverse-V3AE CPU that can handle all computing tasks without an external CPU. This highly integrated design enables robots to achieve true real-time response, making rapid and precise decisions in dynamic environments.
For robot developers, this “brain and cerebellum” integrated design significantly simplifies system architecture, reduces power consumption and size, while improving system reliability and response speed.
1.5 HOLOSCAN Bridging Technology: Revolutionizing High-Speed Data Transmission
Jetson Thor incorporates NVIDIA’s innovative Holoscan sensor bridging technology, which can seamlessly convert various sensor data into network data. This technology provides up to 4x25Gb Ethernet bandwidth, supporting four 25 GbE network interfaces.
1.6 JetPack 7.0: Quantum Leap in Development Experience
Jetson Thor comes with the latest JetPack 7.0 SDK, offering optimized data interfaces and development environment. This system is fully compatible with NVIDIA’s software stack from cloud to edge, including the Isaac platform for robot simulation and development, the Isaac GR00T humanoid robot foundation model, NVIDIA Metropolis for visual AI, and NVIDIA Holoscan for real-time sensor processing.
1.7 Multi-Instance GPU Technology
It enables the partitioning of a single physical GPU into multiple (up to 7) hardware-isolated instances, each equipped with dedicated computing units, memory, and cache resources reserved for specific tasks. This architecture supports parallel and interference-free processing of a robot’s “fast-reaction” and “slow-thinking” tasks, ensuring both operational safety and enhanced intelligent decision-making efficiency.
1.8 The Jetson Thor module is designed to handle lots of video.
Designers can connect up to 20 cameras. It supports video encode up to 6x 4Kp60 for H.265 and H.264 as well as video decode for streams up to 4x 8Kp30 for H.265 and 4x 4Kp60 for H.264. The module also includes NVIDIA’s Programmable Vision Accelerator (PVA). The system can drive up to four HDMI 2.1 or VESA DisplayPort connections.
II. Product Information
2.1 Product Images
（NVIDIA Jetson AGX Thor Developer Kit & Module Official Images）
(NVIDIA Jetson AGX Thor Developer Kit With Box 1)
(Size comparison：NVIDIA Jetson AGX Orin vs AGX Thor Developer kit )
2.2 Product Specifications
| Jetson AGX Thor Developer Kit | Jetson T5000 Module | |
| AI Performance | 2070 TFLOPS (FP4—Sparse) | |
| GPU | 2560-core NVIDIA Blackwell architecture GPU with 96 fifth-gen Tensor Cores | |
| Multi-Instance GPU (MIG) with 10 TPCs | ||
| GPU Max Frequency | 1.57 GHz | |
| CPU | 14-core Arm® Neoverse®-V3AE 64-bit CPU | |
| 1 MB L2 cache per core | ||
| 16 MB shared system L3 cache | ||
| CPU Max Frequency | 2.6 GHz | |
| Vision Accelerator | 1 个 PVA v3 | |
| Memory | 128 GB 256-bit LPDDR5X | |
| 273 GB/s | ||
| Storage | 1 TB NVMe M.2 Key M Slot | Supports NVMe through PCIe |
| Supports SSD through USB3.2 | ||
| Video Encode | 6x 4Kp60 (H.265) | |
| 12x 4Kp30 (H.265) | ||
| 24x 1080p60 (H.265) | ||
| 50x 1080p30 (H.265) | ||
| 48x 1080p30 (H.264) | ||
| 6x 4Kp60 (H.264) | ||
| Video Decode | 4x 8Kp30 (H.265) | |
| 10x 4Kp60 (H.265) | ||
| 22x 4Kp30 (H.265) | ||
| 46x 1080p60 (H.265) | ||
| 92x 1080p30 (H.265) | ||
| 82x 1080p30 (H.264) | ||
| 4x 4Kp60 (H.264) | ||
| Camera | HSB camera via QSFP slot | Up to 20 cameras via HSB |
| USB camera | Up to 6 cameras through 16x lanes MIPI CSI-2 | |
| Up to 32 cameras using Virtual Channels | ||
| C-PHY 2.1 (10.25 Gbps) | ||
| D-PHY 2.1 (40 Gbps) | ||
| PCIe* | M.2 Key M slot with x4 PCIe Gen5 | Up to Gen5 (x8 lanes) |
| M.2 Key E slot with x1 PCIe Gen5 | Root port only—C1 (x1) and C3 (x2) | |
| Root Point or Endpoint—C2 (x1), C4 (x8), and C5 (x4) | ||
| USB* | 2x USB-A (3.2 Gen2) | xHCI host controller with integrated PHY (up to) |
| 2x USB-C (3.1) | 3x USB 3.2 | |
| 4x USB 2.0 | ||
| Networking* | 1x 5GBe RJ45 connector | 4x 25 GbE |
| 1x QSFP28 (4x 25 GbE) | ||
| Display | 1x HDMI 2.0b | 4x shared HDMI2.1 |
| 1x DisplayPort 1.4a | VESA DisplayPort 1.4a—HBR2, MST | |
| Other I/O | QSFP connector | 5x I2S/2x audio hub (AHUB), 2x DMIS, 4x UART, 4x CAN, 3x SPI, 13x I2C, 6x PWM outputs |
| M.2 Key E expansion slot (WLAN/BT, x1 PCIe, USB2.0, UART, I2C, I2S) | ||
| M.2 Key M connector (NVMe for storage) | ||
| PCIe x4 lane, I2C, PCIe x2 lane | ||
| 2x 13-pin CAN header | ||
| 2x 6-pin automation header | ||
| LED | ||
| JTAG connector (2x 5-pin header) | ||
| 1x fan connector —12V, PWM, and Tach | ||
| Audio panel header (2x 5-pin) | ||
| Microfit power jack | ||
| RTC backup battery connector 2-pin | ||
| Power | 40W – 130W | |
| Mechanical | 243.19 mm x 112.40 mm x 56.88 mm | 100 mm x 87 mm |
| Thermal Transfer Plate (TTP) and optional fan or heat sink | 699-pin B2B connector | |
| Integrated Thermal Transfer Plate (TTP) with heatpipe | ||
NVIDIA Jetson AGX Thor T5000 Block Diagram
The Jetson Thor module includes an Arm Neorverse-V3AE CPU cluster and a Blackwell GPU cluster.
Download:
NVIDIA Jetson Thor Module Specifications
NVIDIA Jetson Thor Developer Kit Specifications
2.3 NVIDIA Jetson Thor T5000 Versus Jetson Orin Performance
NVIDIA Jetson Thor T5000 Versus Jetson Orin Performance
Source:https://www.servethehome.com/nvidia-jetson-agx-thor-developer-kit-blackwell-for-robotics/3/
More parameter comparisons:NVIDIA Orin vs Thor: The Generational Leap in Edge AI Computing
2.4 NVIDIA Jetson AGX Thor Product List
1）NVIDIA Jetson T5000 module with heat sink and reference carrier board
2）DC power 140W
3）802.11ax wireless networkinterface controller
4）1 TB NVMe populated in M.2 Key-M slot
5）Quick Start Guide
2.5 NVIDIA Thor Official Video
2.6 NVIDIA Thor Developer Kit Unboxing video
2.7 Product Price and Purchase
Module：
Jetson Thor Module T5000：USD2999
Jetson Thor Module T4000：USD1999
Developer Kit
Jetson Thor Developer Kit：USD3499
If the purchase quantity exceeds 1000pcs, you can contact NVIDIA directly to purchase. If you need less, you can purchase from NVIDIA’s partners.
NVIDIA NPN Partner TWOWIN Technology purchase address：
Jetson Thor Module Specifications 
NVIDIA Jetson Thor Series Modules Design Guide
Jetson Thor Developer Kit Specifications
Jetson Thor Developer Kit User Guide
III. Application Scenarios
3.1 Robotics Industry
Based on information from international sources, NVIDIA’s Jetson Thor computing platform—designed specifically for robotics and physical AI—has been adopted or is under evaluation by several leading companies worldwide. These companies operate across various fields, including humanoid robotics, logistics, healthcare, and agriculture.
Below is a summary table of key international companies and their applications:
| Company Name | Industry | Application/Product | Status |
|---|---|---|---|
| Agility Robotics | Humanoid Robotics | Next-generation Digit humanoid robots for warehouse logistics | Early Adopter |
| Amazon Robotics | Logistics | Logistics robotics systems | Early Adopter |
| Boston Dynamics | Humanoid Robotics | Atlas humanoid robot | Early Adopter |
| Caterpillar | Heavy Machinery | Automation in construction machinery | Early Adopter |
| Figure | Humanoid Robotics | Real-time learning and interaction for humanoid robots | Early Adopter |
| Hexagon | Industrial Automation | Specific application not detailed | Early Adopter |
| Medtronic | Medical Technology | Surgical robots or medical devices | Early Adopter |
| Meta | Technology | Potential AI research applications | Early Adopter |
| 1X (Halodi Robotics) | Humanoid Robotics | Evaluating Thor for advancing physical AI capabilities in robots like NEO | Evaluating |
| John Deere | Agricultural Machinery | Evaluating Thor for smart tractors and agricultural automation systems | Evaluating |
| OpenAI | AI Research | Evaluating Thor for advancing physical AI capabilities | Evaluating |
| Physical Intelligence | Embodied AI Research | Evaluating Thor for advancing physical AI capabilities | Evaluating |
Core Advantages of Jetson Thor
NVIDIA Jetson Thor is highly regarded due to its exceptional performance:
- Impressive Computing Power: Featuring a Blackwell architecture-based GPU, it delivers up to 2070 FP4 TOPS of AI performance, enabling simultaneous operation of multiple complex generative AI models (e.g., large language models, vision-language models).
- Significant Energy Efficiency Improvement: Compared to its predecessor, Jetson Orin, it offers 7.5x higher AI performance and 3.5x better energy efficiency, with a power consumption of just 130 watts.
- Powerful Hardware Configuration: Equipped with 128GB of memory to support complex AI models.
- Addressing Key Challenges: Enables real-time inference at the edge, allowing robots to interact intelligently and in real time with humans and the physical world. This addresses long-standing challenges in robotics, such as computational limitations and latency.
Ecosystem and Impact
NVIDIA’s robotics technology stack has attracted over 2 million developers worldwide. The introduction of Jetson Thor, combined with a comprehensive software suite (including Isaac Sim for simulation and the GR00T humanoid robot foundation model), aims to lower the barrier to robotics development and drive progress across the physical AI and general-purpose robotics industries.
Many companies are currently evaluating how to leverage its capabilities, and more may join the list of adopters in the future.
Source:
https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Blackwell-Powered-Jetson-Thor-Now-Available-Accelerating-the-Age-of-General-Robotics/default.aspx
https://www.marketscreener.com/news/nvidia-blackwell-powered-jetson-thor-now-available-accelerating-the-age-of-general-robotics-ce7c50d8db81f021
3.2 Generative AI
With its Blackwell architecture GPU, AI computing power of up to 2070 FP4 TFLOPS, and 128GB of unified memory, Jetson Thor can efficiently run various large generative AI models directly on the device (edge side). This addresses issues such as high latency, data security, and network stability associated with cloud dependency, making it particularly suitable for physical AI applications that require high real-time performance.
The core of generative AI applications on Jetson Thor lies in enabling machines to “create content” and interact naturally with humans and the environment. Key scenarios include:
- Natural Language Interaction and Decision-Making
- Scenario Description: Robots can engage in natural language conversations with you based on visual information, understand ambiguous instructions, and autonomously plan task steps. For example, if you say, “Clean up the drinks and snacks on the table,” it can identify “drinks” and “snacks” and plan the path and actions for picking and placing them.
- Technical Support: Running large language models (LLMs) and visual language models (VLMs) for real-time multi-turn conversations, complex instruction parsing, and task planning.
- Action Generation and Skill Learning
- Scenario Description: After observing a few human demonstrations, robots can autonomously generate the motion trajectories required to complete similar tasks. Alternatively, when encountering unfamiliar scenes or objects, they can flexibly adjust their strategies.
- Technical Support: Running models like NVIDIA Isaac GR00T, which are visual-language-action models (VLAs), for action generation and optimization.
- Scene Understanding and Simulation
- Scenario Description: Generative AI can be used to create synthetic data needed for training robots or to conduct realistic testing and iteration in their “digital twins.” This primarily relies on cloud-based NVIDIA Omniverse and DGX systems. Jetson Thor is responsible for executing strategies validated through simulation in the real world.
- Multimodal Information Fusion and Generation
- Scenario Description: Robots can simultaneously process data from cameras, microphones, force sensors, and other sources to form a unified understanding of the environment and generate corresponding responses.
- Technical Support: Coordinating and running multiple AI models (e.g., vision, speech, control models) simultaneously for joint reasoning of multimodal data.
Many leading robotics companies and AI research institutions, both domestically and internationally, have begun adopting Jetson Thor to advance their generative AI applications.
| Company Name | Primary Field | Expected/Existing Applications Based on Thor |
| Boston Dynamics | Humanoid Robotics, Dynamic Motion | Providing a more powerful local “brain” for its Atlas humanoid robot to enable more complex autonomous decision-making and interaction. |
| Agility Robotics | Logistics Robotics | Planned for use in the sixth-generation Digit robot to optimize autonomous (handling) and stacking tasks in warehouses. |
| Figure AI | Humanoid Robotics | Partnering with companies like BMW to develop humanoid robots for industrial scenarios, expected to utilize Thor for language interaction and task execution. |
| 1X Technologies | General-Purpose Robotics | Developing robots like NEO, focused on assisting in human environments, using generative AI for learning and adaptation. |
| Amazon Robotics | Logistics Automation | Used in warehousing and logistics robots to enhance autonomous sorting, navigation, and inventory management capabilities. |
| Medtronic | Medical Technology | Exploring applications in surgical assistant robots, potentially for real-time intraoperative guidance and precise operation. |
| Ubtech | Humanoid Robotics | Announced its use in the Walker S2 industrial humanoid robot to enhance multi-robot collaboration and interaction capabilities in smart manufacturing scenarios. |
| Unitree | High-Performance Robotics | Utilizing Thor to improve its robots’ agility, decision-making speed, and autonomous navigation capabilities. |
| Galaxy General | Embodied AI | After adopting Thor, its G1 Premium robot is said to have significantly improved in motion speed and action fluency. |
| Agibot | Humanoid Robotics | Focused on embodied AI, planning to achieve tens of thousands of units shipped next year, with Thor providing core computing support. |
| United Imaging | Medical Equipment | Exploring the application of ultra-high-performance computing in high-end medical equipment, potentially involving medical imaging and AI analysis. |
3.3 Autonomous driving
NVIDIA DRIVE AGX Thor, with its exceptional performance, scalability, and powerful ecosystem, has attracted numerous well-known foreign automakers, autonomous driving technology companies, and Tier 1 suppliers. It is providing critical support for the research, development, and mass-production application of global autonomous driving technologies, driving various vehicle types—including passenger cars, freight trucks, and low-speed delivery vehicles—toward higher levels of automation.
Below is a table summarizing the key foreign companies and their application areas:
| Company Name | Country/Region | Primary Application Area/Role | Remarks/Known Progress |
|---|---|---|---|
| Volvo | Sweden | High-Level Autonomous Driving R&D for Passenger Vehicles (L2+ to L4) | Confirmed adoption of the Thor platform for R&D |
| Aurora | United States | Autonomous Truck R&D | Focuses on L4 autonomous driving for long-haul freight |
| Gatik | United States | Autonomous Medium Truck R&D (Short-Distance Logistics) | Specializes in middle-mile logistics |
| PlusAI | United States | Autonomous Truck Solutions | Provides autonomous driving systems and solutions |
| Waabi | Canada | Autonomous Truck R&D (AI-Driven Approach) | Utilizes deep learning and end-to-end differentiable architectures for innovative R&D |
| Continental | Germany | Tier 1 Supplier, integrating Thor into mass-production systems | Supplies Thor-based autonomous driving domain controllers and other systems to OEMs |
| Magna | Canada/Austria | Tier 1 Supplier, integrating Thor into mass-production systems | Offers full-vehicle engineering and manufacturing capabilities, supplying systems and acting as a contract manufacturer for multiple OEMs |
| Nuro | United States | Low-Speed Autonomous Delivery Vehicles | Focuses on last-mile goods delivery |
| WeRide | China (Headquarters) | L4 Autonomous Robotaxis | Its GXR model is claimed to be the first mass-produced L4 robotaxi, utilizing Thor’s HPC 3.0 system |
Core Advantages of NVIDIA DRIVE AGX Thor
The widespread adoption of the Thor platform is largely due to its following strengths:
- Impressive Computing Power: 2000 TOPS of AI computing power (top-tier dual-chip version), approximately 20 times that of the previous-generation Orin chip. This enables real-time processing of data from multiple sensors, including 14 eight-megapixel cameras, LiDAR, and millimeter-wave radar, to perform complex tasks such as environmental perception, path planning, and decision-making control.
- Advanced Architecture: Utilizes the Blackwell architecture GPU and Arm Neoverse V3AE server-level CPU. This architecture is optimized for emerging workloads like generative AI and visual language models.
- High Scalability and Modular Design: Supports flexible configuration for autonomous driving functions ranging from L2+ to L4 levels. Automakers can rapidly iterate different levels of autonomous driving functionalities based on the same hardware platform.
- Robust Safety and Comprehensive Ecosystem: Complies with stringent automotive industry standards for functional safety (ISO 26262) and cybersecurity (ISO 21434). Its NVIDIA DriveOS 7 software stack is also ISO 26262 certified. Thor is part of the NVIDIA Halos safety framework, which combines hardware, software security technologies, and AI research to provide end-to-end comprehensive safety solutions. From cloud-based DGX (AI training) and Omniverse (simulation) to vehicle-side DRIVE AGX (deployment), NVIDIA offers a complete development toolchain and a closed-loop ecosystem.
Application Trends and Outlook
Current trends in the application of the Thor platform include:
- Parallel Development in Passenger and Commercial Vehicles: Used both to enhance the smart driving experience in passenger vehicles and widely applied in commercial scenarios such as trucks and delivery vehicles.
- Collaboration Between “Full-Stack” and “Supply”: Attracts not only autonomous driving algorithm companies but also integrates key Tier 1 suppliers to jointly advance the maturity of the industry chain.
- L4 High-Level Autonomous Driving as a Key Goal: Many partners aim to achieve the commercialization of L4 unmanned driving.
IV.Stock Price Impact
On August 25, 2025, Nvidia’s stock price was trading at $181.48 (+1.96%), ahead of second-quarter earnings.
Nvidia’s stock has consistently outperformed its benchmark. Year-to-date returns have been 35.27%, compared to the S&P 500’s 9.89%. Over the past year, the stock price has risen 40.43%, compared to the S&P 500’s 14.71%. The company’s long-term performance is even more impressive, with a three-year return of 915.36% and a five-year return of 1,329.17%, far exceeding the S&P 500’s 53.93% and 87.70% returns respectively. With artificial intelligence and robotics at the core of Nvidia’s strategy, its foray into emerging markets like robotics could further solidify its growth trajectory.
Tallu points to the 2.2 million developers across 7,000 companies that have engaged with Jetson – numbers that would have been unheard of in robotics when the platform launched a decade ago.
Source:
https://www.automate.org/industry-insights/nvidias-jetson-thor-robotics-computer-is-now-available
For investors, the case for NVIDIA is clear. The company is not just selling chips—it is building the infrastructure for a $150 billion robotics market. Its ecosystem-driven approach creates switching costs for partners and developers, ensuring long-term lock-in. Meanwhile, the integration of AI into edge computing (via platforms like Holoscan and Metropolis) positions NVIDIA to capture growth in sectors like healthcare, logistics, and smart cities.
NVIDIA’s Jetson Thor is not just a product—it is a strategic inflection point in the AI and robotics landscape. By combining hardware innovation, software integration, and a thriving ecosystem, NVIDIA is setting the standard for physical AI. For investors, this represents a compelling opportunity to ride the next wave of technological disruption. As the robotics market scales, so too will NVIDIA’s influence—and its returns.
V.FAQ
1.NVIDIA Thor vs DGX Spark
Pls check：Comprehensive Comparison of NVIDIA Jetson AGX Thor vs. DGX Spark
2.How does Jetson Thor compare to cloud-based AI processing?
Answer: Thor eliminates cloud latency, reduces bandwidth costs, and enables real-time processing. For robots requiring response times below 10 milliseconds, local processing is essential.
3.Can Thor run large language models locally?
Answer: Yes, with 128GB of RAM, Thor can run models like Llama 70B locally, though inference speed depends on optimization and quantization.
4.Is Thor overkill for simple robotics projects?
Answer: Absolutely. For basic projects, Jetson Nano or Orin Nano are more suitable and cost-effective.
5.What is the learning curve like for Thor development?
Answer: If you are familiar with CUDA and NVIDIA’s ecosystem, the learning process is straightforward. Complete beginners will face a steep learning curve.
6.When will Thor be fully available on the market?
Answer: The developer kit was released on August 25.
7.What are the shortcomings of NVIDIA Thor?
Answer: The device’s performance in running large LLMs may not be ideal, with a bandwidth of approximately 273GB/s (comparable to Strix Halo), while JAO’s MBW only increased by 33%. Orin already delivers 85 FP16 Tensor TFLOPS. An 8x speed boost would result in 680 FP16 TFLOPS (close to most of H100’s performance).
8.Is NVIDIA Thor suitable for individual developers?
The Jetson AGX Thor Developer Kit illustrates Nvidia’s strategic focus on owning the high-performance segment of the embedded AI and robotics market. At $3,499, it is clearly not priced for small-scale tinkerers. Instead, it is targeted at research institutions, robotics startups and enterprises pursuing advanced automation projects.
The NVIDIA Jetson AGX Thor developer kit is now available, priced at $3,499. The Jetson T5000 production modules and Thor developer kit can be purchased through TWOWIN Technology.Twowin is NVIDIA NPN Partner, 11 years of industry experience.
NVIDIA Thor Developer Kit has arrived, please click here for consultation
Other Source:
https://blogs.nvidia.com/blog/jetson-thor-physical-ai-edge/
https://www.cnbc.com/2025/08/25/nvidias-thor-t5000-robot-brain-chip.html
