Jetson Thor: Bringing More Intelligence to the Physical World

At GTC 2025, NVIDIA unveiled Jetson Thor, delivering up to 2,070 FP4 TFLOPS and 128 GB of high-speed memory — a huge leap from the previous 275 FP4 TFLOPS limit. It’s the first Jetson platform capable of running massive AI models, fusing rich sensor data in real time, and still fitting within the power and space constraints of real machines.

Launched now, as humanoid robotics, generative AI, and autonomous systems are accelerating, Thor arrives right when the industry needs it most.

In this article, we’ll explore what Jetson Thor is, why it matters, the kinds of applications it enables, how it compares to previous Jetson generation, and how Forecr’s new DSBOX-THRMAX and DSBOARD-THRMAX platforms make it ready for deployment today.

What Jetson Thor Is

Jetson Thor is NVIDIA’s most advanced edge AI platform to date. It delivers up to 2,070 FP4 TFLOPS of AI compute and supports 128 GB LPDDR5X memory.

It uses the new Blackwell GPU architecture, complete with a Transformer Engine and Multi-Instance GPU (MIG) support. This allows multiple AI models to run at the same time, each with dedicated resources.

Thor is designed for systems that need fast, local decision-making — robots, autonomous machines, industrial automation, and sensor-rich AI devices.

Why It Matters

Runs complex AI at the edge — including large transformer models, generative AI, and advanced perception tasks.
Handles massive data flows with 273 GB/s memory bandwidth, making it ideal for multi-camera, multi-sensor systems.
Efficient performance — packs supercomputer-level compute into a power range of 40–130 W.
Supports real-time control — essential for humanoid robotics, automated manufacturing, and high-speed vision processing.

Jetson Thor vs. Jetson AGX Orin

Thor isn’t just faster. It supports entirely new classes of workloads that earlier Jetson models couldn’t handle in real time.

See the full spec list from NVIDIA!

How It Fits Into Physical AI

Physical AI is AI that interacts with the real world — moving, sensing, reacting. For that to work well, systems need to process data where it’s collected, without delay. Jetson Thor’s compute power, high-speed memory, and ability to run multiple AI models at once make it suited for:

Humanoid robots
Autonomous mobile robots (AMRs)
Smart manufacturing cells
Energy-efficient AI appliances
Multi-sensor inspection systems

Forecr’s DSBOARD-THRMAX & DSBOX-THRMAX

To use Jetson Thor in real-world projects, you need hardware that exposes its full potential. That’s where Forecr’s new platforms come in — available in September.

NVIDIA® Jetson Thor™ Module Industrial Box PC - DSBOX-THRMAX

(Available in September)

A robust industrial box PC built around Jetson Thor, supporting multiple camera inputs, PCIe Gen5 storage, high-speed networking, and a wide operating temperature range. Ready for deployment in harsh field environments.

Engineered to run complex large language models alongside high-throughput video analytics, the DSBOX-THRMAX enables advanced multi-modal AI workloads at the edge — seamlessly integrating real-time visual perception, contextual understanding, and decision-making into a single, deployable platform.

Featuring ultra-fast connectivity via QSFP28 (4x 25 GbE), the DSBOX-THRMAX is built to move massive data streams effortlessly from multi-camera vision to real-time sensor fusion even in extreme -25°C to +85°C conditions. Optional 8x GMSL2 camera inputs and advanced high-speed I/O make it the ultimate platform for demanding multi-sensor AI at the edge.

NVIDIA® Jetson Thor™ Module Carrier Board - DSBOARD-THRMAX

(Available in September)

A compact carrier board for the Jetson T5000 module, with a powerful 128GB LPDDR5X memory interface, high-speed PCIe Gen5 storage, and a wide array of I/O options including dual HDMI 2.1, dual DP1.4a, USB 3.2, QSFP+, and robust serial connectivity. This compact 140x125mm board is engineered to support the most demanding edge AI applications and perfect for integration into robotics, vision systems, and industrial controllers.

Built for seamless deployment in mission-critical environments, the DSBOARD-THRMAX integrates industrial - grade features including an operating range of -25°C to +85°C, ruggedized digital I/Os, CAN bus, and real-time sensor interfaces such as IMU and temperature monitoring. Its flexible expansion ecosystem supports WiFi, Bluetooth, LTE/5G modules, and high-bandwidth camera connectivity, while NVIDIA JetPack 7.0 and Ubuntu Linux 24.04 ensure optimal software compatibility.

Looking Ahead

Jetson Thor makes it possible to bring advanced AI into places where milliseconds matter — factory floors, moving vehicles, outdoor robots. With platforms like DSBOARD-THRMAX and DSBOX-THRMAX, that power becomes easier to build into the systems shaping our future. If you’re ready to explore what’s possible, here’s where to start:

Explore DSBOARD-THRMAX NVIDIA® Jetson Thor™ Module Carrier Board

Explore DSBOX-THRMAX NVIDIA® Jetson Thor™ Module Industrial Box PC