NVIDIA Jetson Thor: Advancing Edge AI Beyond AGX Orin
The launch of NVIDIA Jetson Thor marks a defining moment in the evolution of embedded AI. For the first time, developers can harness server-class performance in a compact system-on-module built for deployment at the edge. It represents not just another hardware release, but a platform that reimagines what is possible for robotics, automation, and intelligent machines.
Jetson Thor combines a 14 core Arm Neoverse V3AE CPU with the NVIDIA Blackwell GPU architecture and its integrated transformer engine, delivering unprecedented performance for edge computing. Paired with 128 GB of high-speed memory, Thor achieves up to 2,070 FP4 TFLOPS of AI performance while scaling power consumption between 40 and 130 watts.
When compared with NVIDIA Jetson AGX Orin, Jetson AGX Thor delivers a dramatic leap forward in performance and efficiency:
7.5× more AI compute, enabling larger models and simultaneous perception pipelines.
3.5× greater efficiency in AI TFLOPs per watt, powering complex multi sensor processing without increasing power consumption.
2.6× faster 14 core Arm Neoverse CPU, accelerating pre-processing and system orchestration.
128 GB LPDDR5X memory on a 256 bit interface, providing 273 GB/s bandwidth for high throughput data handling.
Four 25 GbE interfaces, supporting high bandwidth sensor capture and real time fusion.
Dual video engines, processing up to 4× more 1080p30 H.265 streams for advanced multi camera setups.
NVIDIA Jetson Thor runs on JetPack SDK 7, the most advanced Jetson software release to date. Built on Ubuntu 24.04, it introduces CUDA 13 for next generation GPU acceleration and TensorRT 10 for highly optimised inference, alongside a PREEMPT RT real time kernel and integrated support for Triton inference servers. This combination provides developers with a complete, production ready software stack that is tuned for performance, reliability, and security at the edge.
How Jetson Thor is Set to Transform Edge AI
Jetson AGX Thor redefines what can be achieved at the industrial edge. By consolidating advanced perception, sensor fusion, and AI reasoning onto a single system on module, it eliminates the need for external servers and complex multi-module setups. Robots and autonomous machines can now process more sensors in parallel, make split-second decisions, and interact with their environments with intelligence and precision that was once beyond reach.
“Jetson Orin has already proven itself as a reliable platform for industrial edge AI, but Jetson AGX Thor takes this to an entirely new level. With dramatically higher compute performance, efficiency, and sensor processing capability, it enables applications in manufacturing, logistics, humanoids, and smart infrastructure that were previously out of reach. By working closely with our ecosystem of NVIDIA Jetson and vision camera partners, Things Embedded is ready to help customers deploy Jetson Thor in real-world environments.” Andrew Whitehouse, Managing Director at Things Embedded.
Humanoids
Advanced humanoids require vision for locomotion, recognition, and dexterous manipulation all at once. Depth sensing combined with Thor’s MIG partitioning keeps these pipelines running in parallel, enabling safe and natural collaboration with people.
Manufacturing
Robots can handle inspection, bin picking, quality control, and worker safety simultaneously on a single module. With Thor’s expanded compute and video throughput, they remain accurate and responsive, driving higher productivity without compromise.
Logistics and Warehousing
Autonomous mobile robots and forklifts gain wide field 3D vision for aisle navigation, pallet handling, and dock operations. Jetson Thor’s efficiency lets fleets scale sensor counts while keeping mapping and obstacle avoidance sharp and responsive.
Smart Facilities and Infrastructure
Safety monitoring, flow analytics, and asset tracking can be consolidated onto a single system. By reducing hardware sprawl across large sites, Jetson Thor lowers cost while enabling adaptive, intelligent facilities that respond in real time.
Sensor Fusion and Real-Time Intelligence at the Edge
Perception is the foundation of autonomy, and Jetson Thor is engineered to transform raw sensor data into actionable intelligence in real time. By uniting performance, efficiency, and reliability in a single platform, machines can perceive, reason, and act with unprecedented clarity and speed.
A cornerstone of Jetson Thor’s architecture is NVIDIA Multi Instance GPU (MIG) technology, which allows a single GPU to be divided into multiple isolated instances. This makes it possible for perception, mapping, and planning workloads to run independently in parallel, each with its own dedicated compute and memory resources. By preventing resource contention, MIG ensures that performance remains consistent and predictable.
Combining the Jetson Thor’s massive memory bandwidth with four 25 GbE connections gives developers a straightforward way to bring together streams from multiple cameras alongside data from LiDAR and radar. Instead of struggling with bottlenecks, systems can now move huge volumes of sensor data smoothly into real time AI pipelines. Adding to this, NVIDIA Holoscan sensor to GPU pathways keeps information flowing directly into GPU memory, avoiding unnecessary CPU handoffs and cutting latency to a minimum. The result is a platform that delivers faster perception, more reliable sensor fusion, and the responsiveness required for advanced applications.
This leap in capability is already reshaping what autonomy looks like in the real world. Put into perspective, Jetson Thor is not just raising the bar for performance—it is redefining what is possible for autonomy across healthcare, logistics, and advanced manufacturing. In the operating room, surgical robots can process four simultaneous 4K video streams in as little as 11 milliseconds, giving surgeons the clarity and responsiveness needed for delicate, high-stakes procedures. On the move, autonomous mobile robots and vehicles achieve reliable situational awareness by fusing LiDAR, radar, and camera inputs in real time, allowing them to navigate dynamic environments with confidence and precision. On the production line, industrial inspection systems can harness multi-camera vision AI at unprecedented throughput without sacrificing latency, ensuring every product is verified at speed.
From Prototype to Deployment: Industrial-Grade Jetson Thor Edge AI Systems
At Things Embedded, we bridge the gap between prototyping and deployment by taking NVIDIA Jetson technology out of the lab and into the field. Developer kits are invaluable for innovation, but they are not designed for the realities of long-life, mission-critical operation. That’s where we come in.
We design and deliver Jetson Thor carrier boards and box PCs built specifically for harsh environments, supporting wide temperature ranges, resistance to shock and vibration, and dependable 24/7 uptime. Our industrial-grade Edge AI platforms are ruggedised, industry-certified, and purpose-built to integrate NVIDIA Jetson Thor into real-world applications.
By combining Thor’s cutting-edge AI performance with our proven hardware expertise, we enable OEMs and system integrators to move beyond the prototype stage and deploy edge AI with confidence. Whether it’s a compact carrier board for seamless integration into your own assembly or a fully sealed IP-rated computer for outdoor deployment, Things Embedded provides the solutions that make Jetson Thor ready for the environments where it matters most.
Ready to Deploy Jetson Thor in the Real World?
Tell us about your environment and workload, and we’ll help you design a system that lasts.